CRP Plan Report

Versions

• v9 # Changelog ## 6.0.0 The reports agent is the only way the plan is laid out: the agent no longer draws its own design, so there is no design proof, no saved design and no check or keep stage. When the user turns the plans down they are asked what to change -- the wording, through the narrative, or the cover photo -- and all three plans are laid out again. A wording change after the narrative is approved applies on top of the approved text and asks for the narrative's approval again, and the plans are not laid out until it is given. Each plan has a cover photo found once from Unsplash for a subject the outline names, wind turbines in a green landscape when it names none, and shared by all three versions; `--cover-image none` keeps the brand's own cover. A plan for guidelines that are not the organisation's is laid out with `--brand none`. The scenario analysis and the baseline carry real headings, each kept on the page of the text and table it introduces. Designs saved in memory by earlier versions are left in place and no longer read. ## 5.1.0 Section 4 is the scenario analysis, as agreed, under the title "Scenario analysis": the reduction required, what each scope was made of, every initiative with its basis, and the Scope 3 deep dive, its tables as tables. It is read from memory and never written by the model. The plan is made three times, once for each version of the Scope 3 deep dive -- fixed categories, focused and broad -- and the user is told how they differ. A plan asked for through "Generate carbon reduction plan report" raises no card. A scenario whose deep dive is out of date is refused until it is brought up to date, and a section brief that cannot be built fails rather than being reported as the writing service not responding. A share prints with one decimal place, or two or three when fewer would show it as zero, and below that as "<0.001%". The Baseline section shows the footprint at a glance under its text: emissions by scope against the current period, charted and tabled, a composition chart tracing every category, however small, to its share of the footprint, and each scope's categories with their share of it. Each pathway chart is preceded by its commitment's targets. When the reports agent cannot lay the plan out, the plan stage stops with its reason instead of the plan being built another way. A saved design is offered by where it is kept rather than copied into the plan's data, and one accepted automatically is never saved. ## 5.0.0 The plan is laid out by the reports agent first and offered on one approval card; the agent draws its own design from the brand only when the user turns that plan down, and the approved narrative carries over to it. The plan runs as the last step of the carbon reduction run as well as on request. When the user chose automatic sign-off at the start of the run, no card is raised and each acceptance is named in one line. ## 4.5.0 The cover is drawn to one layout: the logo in a white card, an eyebrow, the title in capitals over an accent rule, a wind-and-landscape motif, three headline tiles across the page and a two-column foot. A design approved before this cover is not offered back. A chart is held to the width of its card, so a wide chart no longer shrinks every page and the cover reaches all four page edges. The check stage fails a page whose content runs past its margin, naming the page and the text where it happens, so a line that will not wrap or a page that holds too much no longer shrinks the whole plan. A brand that names no usable palette takes its colours from its logo, most used first, and the plan's primary is the palette's first colour, so the plan wears the organisation's own colours rather than invented ones. ## 4.4.0 A design approved before the current rules is no longer offered back, so an organisation that approved a plain look still gets the plan the rules describe; the design saved on approval records the rules it was drawn to. Body text is set at 11pt, tables at 9.5pt and the cover title at 28pt or more, so the plan reads on paper. ## 4.3.0 The plan is drawn as a published report at the first attempt: a cover whose colour field reaches the page edges, the logo at real size, the headline figures as tiles, and a quiet brand-coloured motif where one earns its place. The print rules that make a full bleed possible are given -- no page margin, a page wrapper carrying the padding -- with the rules that keep every page holding its content: one wrapper per page, the footer as its last child, nothing taller than the page box, and a continued table repeating its header. Colour is taken from the brand to a stated contrast floor, so a dark field never costs legibility. ## 4.2.0 The plan is drawn to the shape of a board document: running header and footer, numbered sections, stat tiles, tables with totals, charts in captioned cards, and a sign-off page. The faults seen in real plans are named so they are not repeated -- raw markdown, broken pages, a missing logo, invented credentials. A design turned down twice can be rendered by the reports agent instead, in its own house style, held to the same checks. An agent call refused because the sandbox predates the blueprint's agents recreates the sandbox once and retries. ## 4.1.1 - The narrative is shown in the reply instead of being named as a file in the outputs. ## 4.1.0 The publisher's mark is sized to the line it sits on, so it no longer spreads across the footer. Every design proof, and the finished plan, is shown in the reply itself, including after a revision. When the brand guidelines never mention the organisation's name, the agent asks once, before designing, whether the plan carries that name with those guidelines. The cover has one clear order: the organisation's name and logo, the title, the period and date. ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 10/8/2026
• v8 # Changelog ## 5.1.0 Section 4 is the scenario analysis, as agreed, under the title "Scenario analysis": the reduction required, what each scope was made of, every initiative with its basis, and the Scope 3 deep dive, its tables as tables. It is read from memory and never written by the model. The plan is made three times, once for each version of the Scope 3 deep dive -- fixed categories, focused and broad -- and the user is told how they differ. A plan asked for through "Generate carbon reduction plan report" raises no card. A scenario whose deep dive is out of date is refused until it is brought up to date, and a section brief that cannot be built fails rather than being reported as the writing service not responding. A share prints with one decimal place, or two or three when fewer would show it as zero, and below that as "<0.001%". The Baseline section shows the footprint at a glance under its text: emissions by scope against the current period, charted and tabled, a composition chart tracing every category, however small, to its share of the footprint, and each scope's categories with their share of it. Each pathway chart is preceded by its commitment's targets. When the reports agent cannot lay the plan out, the plan stage stops with its reason instead of the plan being built another way. A saved design is offered by where it is kept rather than copied into the plan's data, and one accepted automatically is never saved. ## 5.0.0 The plan is laid out by the reports agent first and offered on one approval card; the agent draws its own design from the brand only when the user turns that plan down, and the approved narrative carries over to it. The plan runs as the last step of the carbon reduction run as well as on request. When the user chose automatic sign-off at the start of the run, no card is raised and each acceptance is named in one line. ## 4.5.0 The cover is drawn to one layout: the logo in a white card, an eyebrow, the title in capitals over an accent rule, a wind-and-landscape motif, three headline tiles across the page and a two-column foot. A design approved before this cover is not offered back. A chart is held to the width of its card, so a wide chart no longer shrinks every page and the cover reaches all four page edges. The check stage fails a page whose content runs past its margin, naming the page and the text where it happens, so a line that will not wrap or a page that holds too much no longer shrinks the whole plan. A brand that names no usable palette takes its colours from its logo, most used first, and the plan's primary is the palette's first colour, so the plan wears the organisation's own colours rather than invented ones. ## 4.4.0 A design approved before the current rules is no longer offered back, so an organisation that approved a plain look still gets the plan the rules describe; the design saved on approval records the rules it was drawn to. Body text is set at 11pt, tables at 9.5pt and the cover title at 28pt or more, so the plan reads on paper. ## 4.3.0 The plan is drawn as a published report at the first attempt: a cover whose colour field reaches the page edges, the logo at real size, the headline figures as tiles, and a quiet brand-coloured motif where one earns its place. The print rules that make a full bleed possible are given -- no page margin, a page wrapper carrying the padding -- with the rules that keep every page holding its content: one wrapper per page, the footer as its last child, nothing taller than the page box, and a continued table repeating its header. Colour is taken from the brand to a stated contrast floor, so a dark field never costs legibility. ## 4.2.0 The plan is drawn to the shape of a board document: running header and footer, numbered sections, stat tiles, tables with totals, charts in captioned cards, and a sign-off page. The faults seen in real plans are named so they are not repeated -- raw markdown, broken pages, a missing logo, invented credentials. A design turned down twice can be rendered by the reports agent instead, in its own house style, held to the same checks. An agent call refused because the sandbox predates the blueprint's agents recreates the sandbox once and retries. ## 4.1.1 - The narrative is shown in the reply instead of being named as a file in the outputs. ## 4.1.0 The publisher's mark is sized to the line it sits on, so it no longer spreads across the footer. Every design proof, and the finished plan, is shown in the reply itself, including after a revision. When the brand guidelines never mention the organisation's name, the agent asks once, before designing, whether the plan carries that name with those guidelines. The cover has one clear order: the organisation's name and logo, the title, the period and date. ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 9/29/2026
• v7 # Changelog ## 5.1.0 Section 4 is the scenario analysis, as agreed, under the title "Scenario analysis": the reduction required, what each scope was made of, every initiative with its basis, and the Scope 3 deep dive, its tables as tables. It is read from memory and never written by the model. The plan is made three times, once for each version of the Scope 3 deep dive -- fixed categories, focused and broad -- and the user is told how they differ. A plan asked for through "Generate carbon reduction plan report" raises no card. A scenario whose deep dive is out of date is refused until it is brought up to date, and a section brief that cannot be built fails rather than being reported as the writing service not responding. A share prints with one decimal place, or two or three when fewer would show it as zero, and below that as "<0.001%". The Baseline section shows the footprint at a glance under its text: emissions by scope against the current period, and each scope's categories with their share of it. ## 5.0.0 The plan is laid out by the reports agent first and offered on one approval card; the agent draws its own design from the brand only when the user turns that plan down, and the approved narrative carries over to it. The plan runs as the last step of the carbon reduction run as well as on request. When the user chose automatic sign-off at the start of the run, no card is raised and each acceptance is named in one line. ## 4.5.0 The cover is drawn to one layout: the logo in a white card, an eyebrow, the title in capitals over an accent rule, a wind-and-landscape motif, three headline tiles across the page and a two-column foot. A design approved before this cover is not offered back. A chart is held to the width of its card, so a wide chart no longer shrinks every page and the cover reaches all four page edges. The check stage fails a page whose content runs past its margin, naming the page and the text where it happens, so a line that will not wrap or a page that holds too much no longer shrinks the whole plan. A brand that names no usable palette takes its colours from its logo, most used first, and the plan's primary is the palette's first colour, so the plan wears the organisation's own colours rather than invented ones. ## 4.4.0 A design approved before the current rules is no longer offered back, so an organisation that approved a plain look still gets the plan the rules describe; the design saved on approval records the rules it was drawn to. Body text is set at 11pt, tables at 9.5pt and the cover title at 28pt or more, so the plan reads on paper. ## 4.3.0 The plan is drawn as a published report at the first attempt: a cover whose colour field reaches the page edges, the logo at real size, the headline figures as tiles, and a quiet brand-coloured motif where one earns its place. The print rules that make a full bleed possible are given -- no page margin, a page wrapper carrying the padding -- with the rules that keep every page holding its content: one wrapper per page, the footer as its last child, nothing taller than the page box, and a continued table repeating its header. Colour is taken from the brand to a stated contrast floor, so a dark field never costs legibility. ## 4.2.0 The plan is drawn to the shape of a board document: running header and footer, numbered sections, stat tiles, tables with totals, charts in captioned cards, and a sign-off page. The faults seen in real plans are named so they are not repeated -- raw markdown, broken pages, a missing logo, invented credentials. A design turned down twice can be rendered by the reports agent instead, in its own house style, held to the same checks. An agent call refused because the sandbox predates the blueprint's agents recreates the sandbox once and retries. ## 4.1.1 - The narrative is shown in the reply instead of being named as a file in the outputs. ## 4.1.0 The publisher's mark is sized to the line it sits on, so it no longer spreads across the footer. Every design proof, and the finished plan, is shown in the reply itself, including after a revision. When the brand guidelines never mention the organisation's name, the agent asks once, before designing, whether the plan carries that name with those guidelines. The cover has one clear order: the organisation's name and logo, the title, the period and date. ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 9/29/2026
• v6 # Changelog ## 5.1.0 Section 4 is the scenario analysis, as agreed, under the title "Scenario analysis": the reduction required, what each scope was made of, every initiative with its basis, and the Scope 3 deep dive, its tables as tables. It is read from memory and never written by the model. The plan is made three times, once for each version of the Scope 3 deep dive -- fixed categories, focused and broad -- and the user is told how they differ. A plan asked for through "Generate carbon reduction plan report" raises no card. A scenario whose deep dive is out of date is refused until it is brought up to date, and a section brief that cannot be built fails rather than being reported as the writing service not responding. Shares print to three decimal places. The Baseline section shows the footprint at a glance under its text: emissions by scope against the current period, and each scope's categories with their share of it. ## 5.0.0 The plan is laid out by the reports agent first and offered on one approval card; the agent draws its own design from the brand only when the user turns that plan down, and the approved narrative carries over to it. The plan runs as the last step of the carbon reduction run as well as on request. When the user chose automatic sign-off at the start of the run, no card is raised and each acceptance is named in one line. ## 4.5.0 The cover is drawn to one layout: the logo in a white card, an eyebrow, the title in capitals over an accent rule, a wind-and-landscape motif, three headline tiles across the page and a two-column foot. A design approved before this cover is not offered back. A chart is held to the width of its card, so a wide chart no longer shrinks every page and the cover reaches all four page edges. The check stage fails a page whose content runs past its margin, naming the page and the text where it happens, so a line that will not wrap or a page that holds too much no longer shrinks the whole plan. A brand that names no usable palette takes its colours from its logo, most used first, and the plan's primary is the palette's first colour, so the plan wears the organisation's own colours rather than invented ones. ## 4.4.0 A design approved before the current rules is no longer offered back, so an organisation that approved a plain look still gets the plan the rules describe; the design saved on approval records the rules it was drawn to. Body text is set at 11pt, tables at 9.5pt and the cover title at 28pt or more, so the plan reads on paper. ## 4.3.0 The plan is drawn as a published report at the first attempt: a cover whose colour field reaches the page edges, the logo at real size, the headline figures as tiles, and a quiet brand-coloured motif where one earns its place. The print rules that make a full bleed possible are given -- no page margin, a page wrapper carrying the padding -- with the rules that keep every page holding its content: one wrapper per page, the footer as its last child, nothing taller than the page box, and a continued table repeating its header. Colour is taken from the brand to a stated contrast floor, so a dark field never costs legibility. ## 4.2.0 The plan is drawn to the shape of a board document: running header and footer, numbered sections, stat tiles, tables with totals, charts in captioned cards, and a sign-off page. The faults seen in real plans are named so they are not repeated -- raw markdown, broken pages, a missing logo, invented credentials. A design turned down twice can be rendered by the reports agent instead, in its own house style, held to the same checks. An agent call refused because the sandbox predates the blueprint's agents recreates the sandbox once and retries. ## 4.1.1 - The narrative is shown in the reply instead of being named as a file in the outputs. ## 4.1.0 The publisher's mark is sized to the line it sits on, so it no longer spreads across the footer. Every design proof, and the finished plan, is shown in the reply itself, including after a revision. When the brand guidelines never mention the organisation's name, the agent asks once, before designing, whether the plan carries that name with those guidelines. The cover has one clear order: the organisation's name and logo, the title, the period and date. ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 9/29/2026
• v5 # Changelog ## 5.1.0 Section 4 is the scenario analysis, as agreed, under the title "Scenario analysis": the reduction required, what each scope was made of, every initiative with its basis, and the Scope 3 deep dive, its tables as tables. It is read from memory and never written by the model. A plan asked for through "Generate carbon reduction plan report" raises no card. ## 5.0.0 The plan is laid out by the reports agent first and offered on one approval card; the agent draws its own design from the brand only when the user turns that plan down, and the approved narrative carries over to it. The plan runs as the last step of the carbon reduction run as well as on request. When the user chose automatic sign-off at the start of the run, no card is raised and each acceptance is named in one line. ## 4.5.0 The cover is drawn to one layout: the logo in a white card, an eyebrow, the title in capitals over an accent rule, a wind-and-landscape motif, three headline tiles across the page and a two-column foot. A design approved before this cover is not offered back. A chart is held to the width of its card, so a wide chart no longer shrinks every page and the cover reaches all four page edges. The check stage fails a page whose content runs past its margin, naming the page and the text where it happens, so a line that will not wrap or a page that holds too much no longer shrinks the whole plan. A brand that names no usable palette takes its colours from its logo, most used first, and the plan's primary is the palette's first colour, so the plan wears the organisation's own colours rather than invented ones. ## 4.4.0 A design approved before the current rules is no longer offered back, so an organisation that approved a plain look still gets the plan the rules describe; the design saved on approval records the rules it was drawn to. Body text is set at 11pt, tables at 9.5pt and the cover title at 28pt or more, so the plan reads on paper. ## 4.3.0 The plan is drawn as a published report at the first attempt: a cover whose colour field reaches the page edges, the logo at real size, the headline figures as tiles, and a quiet brand-coloured motif where one earns its place. The print rules that make a full bleed possible are given -- no page margin, a page wrapper carrying the padding -- with the rules that keep every page holding its content: one wrapper per page, the footer as its last child, nothing taller than the page box, and a continued table repeating its header. Colour is taken from the brand to a stated contrast floor, so a dark field never costs legibility. ## 4.2.0 The plan is drawn to the shape of a board document: running header and footer, numbered sections, stat tiles, tables with totals, charts in captioned cards, and a sign-off page. The faults seen in real plans are named so they are not repeated -- raw markdown, broken pages, a missing logo, invented credentials. A design turned down twice can be rendered by the reports agent instead, in its own house style, held to the same checks. An agent call refused because the sandbox predates the blueprint's agents recreates the sandbox once and retries. ## 4.1.1 - The narrative is shown in the reply instead of being named as a file in the outputs. ## 4.1.0 The publisher's mark is sized to the line it sits on, so it no longer spreads across the footer. Every design proof, and the finished plan, is shown in the reply itself, including after a revision. When the brand guidelines never mention the organisation's name, the agent asks once, before designing, whether the plan carries that name with those guidelines. The cover has one clear order: the organisation's name and logo, the title, the period and date. ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 9/28/2026
• v4 # Changelog ## 5.0.0 The plan is laid out by the reports agent first and offered on one approval card; the agent draws its own design from the brand only when the user turns that plan down, and the approved narrative carries over to it. The plan runs as the last step of the carbon reduction run as well as on request. When the user chose automatic sign-off at the start of the run, no card is raised and each acceptance is named in one line. ## 4.5.0 The cover is drawn to one layout: the logo in a white card, an eyebrow, the title in capitals over an accent rule, a wind-and-landscape motif, three headline tiles across the page and a two-column foot. A design approved before this cover is not offered back. A chart is held to the width of its card, so a wide chart no longer shrinks every page and the cover reaches all four page edges. The check stage fails a page whose content runs past its margin, naming the page and the text where it happens, so a line that will not wrap or a page that holds too much no longer shrinks the whole plan. A brand that names no usable palette takes its colours from its logo, most used first, and the plan's primary is the palette's first colour, so the plan wears the organisation's own colours rather than invented ones. ## 4.4.0 A design approved before the current rules is no longer offered back, so an organisation that approved a plain look still gets the plan the rules describe; the design saved on approval records the rules it was drawn to. Body text is set at 11pt, tables at 9.5pt and the cover title at 28pt or more, so the plan reads on paper. ## 4.3.0 The plan is drawn as a published report at the first attempt: a cover whose colour field reaches the page edges, the logo at real size, the headline figures as tiles, and a quiet brand-coloured motif where one earns its place. The print rules that make a full bleed possible are given -- no page margin, a page wrapper carrying the padding -- with the rules that keep every page holding its content: one wrapper per page, the footer as its last child, nothing taller than the page box, and a continued table repeating its header. Colour is taken from the brand to a stated contrast floor, so a dark field never costs legibility. ## 4.2.0 The plan is drawn to the shape of a board document: running header and footer, numbered sections, stat tiles, tables with totals, charts in captioned cards, and a sign-off page. The faults seen in real plans are named so they are not repeated -- raw markdown, broken pages, a missing logo, invented credentials. A design turned down twice can be rendered by the reports agent instead, in its own house style, held to the same checks. An agent call refused because the sandbox predates the blueprint's agents recreates the sandbox once and retries. ## 4.1.1 - The narrative is shown in the reply instead of being named as a file in the outputs. ## 4.1.0 The publisher's mark is sized to the line it sits on, so it no longer spreads across the footer. Every design proof, and the finished plan, is shown in the reply itself, including after a revision. When the brand guidelines never mention the organisation's name, the agent asks once, before designing, whether the plan carries that name with those guidelines. The cover has one clear order: the organisation's name and logo, the title, the period and date. ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 9/28/2026
• v3 # Changelog ## 4.1.1 - The narrative is shown in the reply instead of being named as a file in the outputs. ## 4.1.0 The publisher's mark is sized to the line it sits on, so it no longer spreads across the footer. Every design proof, and the finished plan, is shown in the reply itself, including after a revision. When the brand guidelines never mention the organisation's name, the agent asks once, before designing, whether the plan carries that name with those guidelines. The cover has one clear order: the organisation's name and logo, the title, the period and date. ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 9/24/2026
• v2 # Changelog ## 4.1.0 The publisher's mark is sized to the line it sits on, so it no longer spreads across the footer. Every design proof, and the finished plan, is shown in the reply itself, including after a revision. When the brand guidelines never mention the organisation's name, the agent asks once, before designing, whether the plan carries that name with those guidelines. The cover has one clear order: the organisation's name and logo, the title, the period and date. ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 9/23/2026
• v1 # Changelog ## 4.0.0 The agent designs the plan itself from the organisation's brand guidelines, or keeps the design the user last approved. Pathway charts, one for Scope 1 and 2 and one for Scope 3, and the category charts are drawn by the chart agents. The narrative is still written and checked figure by figure, and the finished PDF is still checked to show every section, task, key figure and the sign-off declaration. Projections are stated without markers. A scenario or pathway agreed before this version keeps the projections it was agreed with until it is agreed again. ## 3.2.2 Changes the user asks for that cannot be made readable keep the design they already reviewed, with its own proof, instead of replacing it with the standard design. The standard design is still what a first design falls back to, and what an organisation with no brand guidelines gets. The publisher's mark is inlined and drawn in the colour of the text around it, so it reads on a dark cover as well as a light one. ## 3.2.1 Every category the plan lists is now drawn. A bar with a value draws at least a minimum length, and the donut folds a tail of slices too thin to tell apart into one "Other categories" slice, widening any remaining slice that would be too narrow to see. The trajectory chart draws coincident pathways as one line named for both, so a commitment at the SBTi floor reads "Target Pathway (= SBTi minimum 4.2%/yr)" rather than two lines a reader cannot separate. A readability problem is numbered by where its verdict falls in the review, not by the page number the review claims. ## 3.2.0 The readability review is its own run (`--stage review`), so a design and its review each have a tool call to themselves. A faulted design is revised against exactly what the review named, keeping what passed, for up to three passes before the standard design is offered. The design instruction asks for chart colours drawn from the brand, stepped in lightness so neighbouring marks are told apart, with a dash pattern per pathway line. The "Prepared with Qonstrue" mark carries its own size and spacing, so it reads the same in a design that never styles it. ## 3.1.2 The design call gets the read timeout a whole document needs (a measured design takes about two minutes, against the broker's 55-second default), so a design is drawn rather than timing out. A design that could not be made at all now says the design service did not answer, instead of reporting that the design failed its checks. ## 3.1.1 The test logo ships base64-encoded, so the skill installs. ## 3.1.0 The logo's colours are read from the logo image itself, with no service call. The standard design takes the logo's main colour for headings, rules, table bands and chart bands, darkened where text on white would not read; it is what a failed design offers instead, and every new design is asked for starting from it. Asking for changes to an offered design, standard or drawn, revises that design. The "Prepared with Qonstrue" mark sits at the size and baseline of the text around it. The scenario's targets table is headed "Emissions reduction targets", with columns for the target, baseline, reduction, target emissions, current emissions and remaining reduction. Targets are named near-term and long-term throughout the plan and its narrative. ## 3.0.0 The plan is made in three stages, each its own run that ends with the user: `--stage design` offers a proof of the design drawn from the brand guidelines and saves it only once approved, or redesigns once with the user's requested changes; `--stage content` offers the five sections' narrative, which the user approves, replaces section by section or asks to be written again; `--stage render` fills the approved design with the approved narrative. A saved design must carry its approval to be reused. A design that fails its checks offers the standard design instead, and a plan the approved design cannot show exits 3 rather than switching design. Refusals exit 4. Every stage fits a 600-second tool call. Every figure prints one way: tonnes with thousands separators to the decimal places the assessed standard sets (2 by default), shares to one decimal place, committed percentages, years and counts whole. The narrative is written from those printed figures, and a section stating a figure to more decimal places than the plan prints uses standard wording. The action plan's id is read from `planId` at the top of the organisation's profile, the tasks directory's `.meta.json` files are skipped, and the PDF text check requires each item at least once. ## 2.1.0 A scenario measured from the baseline is narrated that way: the goalposts show today for reference with nothing "still to remove", the projection legend measures against the baseline, and section 4 says what must change from the baseline and where the delivery pathway falls short, never counting today's emissions as progress. A scenario signed off before it recorded a basis keeps its today-based columns and wording. ## 2.0.0 The plan's design is drawn from the organisation's brand guidelines and logo: a model designs the document once, it is checked, proofed and reviewed for readability, then saved and reused until the brand guidelines change or the user asks for a new look (`--new-look`). A design that does not pass, or no brand guidelines, gives the standard design. The plan gains a contents page, a sign-off page, scope and commitment cards, category bar and donut charts, the Scope 1 and 2 breakdown and the Scope 3 products and suppliers behind the scenario, and completed and planned projects. Each section's narrative is written from its own brief and data, and section 3 draws on the registry clauses the assessment cites. Every rendered PDF is checked to show every task and key figure before it is handed over. ## 1.0.0 Added the skill: `report_build.py` composes an organisation's already-agreed figures, scenario and action plan into a Carbon Reduction Plan -- a cover page, contents, and five sections (executive summary, baseline, SBTi criteria and targets, the least ambitious scenario, and the action plan and initiatives) -- and converts it to one paged, branded PDF. The brand guidelines supply the logo and the accent colour only, never typefaces. Each section's narrative is model-written and may state only figures the model was given from the organisation's own data; a section whose written narrative cites any other figure uses standard wording instead, and the user is told which. A run refuses to produce a plan missing any of the organisation's action-plan tasks. 9/22/2026