Session date: <yyyy-mm-dd> Reply with the numbers to drop, "scores only" for no quotes, or any change to the title or date. ``` - **Quotes.** Choose only from the Evidence quotes the assessment already showed, for capabilities that have a level. At most 5, each at most 240 characters, each exactly as the user wrote it (check against the assessed chat; drop any you can't match). Keep one per scored capability first, in rubric order, then add second quotes in the same order until you have 5. Show the capability each one is tied to. Never include a quote for a "not shown" capability, and never a quote from the assistant's replies. - **Title** (at most 120 characters): the chat's own title when the assessment named one (cut to fit), otherwise a plain description of the subject in a few words. Leave out names of people, companies, email addresses and amounts. - **Date**: the day the chat took place as `yyyy-mm-dd`: the date shown for it in a chat export, past chat or session file; for a chat that was pasted or is the current one and shows no date, today's date if your context gives it. If you have no date at all, ask for one in the same message. - If you already showed a title and date earlier in this conversation (for example in an export plan), use exactly those in the report and record unless the user changed them. Don't swap in today's date or comment on the date. ## 3. The user decides - They can drop quotes by number, keep all, say "scores only" (the evidence list is then empty and no quote appears anywhere in the report), or change the title or date. Apply exactly what they say. A dropped quote appears nowhere in the output. - If they reply with a change you can't apply (a quote of their own that is not in the chat, or a title over 120 characters), say so in one line and keep the proposal. Don't add quotes the user didn't approve, even if they ask for "everything relevant"; propose them back for approval. - A request to export that already says what to include (for example "scores only") can skip straight to step 4. ## 4. Show exactly what will be in the file Write the complete report below, exactly as it would be saved, so the user sees every character. Use this layout (the outer fence here is four backticks only because the report contains its own fenced block): ```` # Modo AI Skills assessment: <session_title> Chat date: <session_date>. Rubric: v1.1. Based on 1 chat. Self-assessment by ChatGPT using Modo rubric v1.1 — not verified. | Capability | Level | Confidence | |---|---|---| | Problem Solving | 4 of 5 | medium | | Verification | 4 of 5 | high | | Workflow Design | 2 of 5 | medium | | Delegation | 1 of 5 | medium | | Strategic Thinking | Not shown | low (see note) | Levels run from 1 to 5. "Not shown" means nothing in the chat bears on that capability. Confidence is low, medium or high, as defined in the rubric. The table is in rubric order. This is for your own development, not for hiring or other employment decisions. Note: for a capability that is not shown, the record gives confidence "low", because no moment in the chat supports a level. ## Evidence you approved - Verification: "<quote>" - ... (Or, for a scores-only export: "None. This export has scores only.") ## Assessment record (machine-readable) ```json { ...the record... } ``` ```` - The record in the final block follows RECORD.md exactly: keys `rubric_version` (`"v1.1"`), `capabilities` (the five keys in rubric order), `evidence`, `session_title`, `session_date`, in that order, with no other fields. Use `"not_shown"` for a capability with no level and give it confidence `"low"`. Print it pretty (2-space indent). - The table, the evidence list and the block must agree. The evidence list shows the same quotes, in the same order, as the block's `evidence`. - The block is the **last thing in the report**, and the report has no other `json` block. Nothing follows it in the file. - Don't put the report inside a code fence of its own; the nested fence would break the block. - Keep the table, headings and wording as above so the report reads on its own without the chat. - Before sending, check the limits against RECORD.md: at most 5 quotes, each 1 to 240 characters, title 1 to 120 characters, date a real `yyyy-mm-dd`, exactly five capabilities, levels 1 to 5 or `"not_shown"`, a confidence on every capability. ## 5. Write the file, only when asked Put one short line **before** the report (the record block must be the last thing in your message, with at most the label after it), depending on your tools. Decide this from the tools you actually have in this session: you can create files only if one of them writes files (a file-writing or file-creation tool, or a code tool that can create a file the user downloads). Reading and searching tools don't count. If you are not sure you have one, you don't, and you must not offer to save a file or a PDF: - **You can create files here.** "This is exactly what would be saved as `modo-assessment-<session_date>.md`. Say 'save it' (or give another name) and I will; nothing is written until then." Create the file, with exactly the text shown, only after the user says so. Then give its download link or path. Don't save anywhere else and don't save a second copy. - **You can't create files here** (a plain chat with no file creation, for example): "I can't save files in this session, so this report is the export: copy it, record block last, into a `.md` file." Don't offer to upload it. If they ask where to upload it: uploading is something they do themselves; which file the Modo upload page accepts is not specified in this plugin. ## 6. PDF, only when asked Make a PDF only if the user asks for one. Show the same Markdown report first (step 4), with the same content in the same order and the record block last. Then create the PDF with whatever file creation you have in this session (in ChatGPT, usually its code tool) and the user's approval. Give the download link or say where it is. The template belongs to this skill, so a code tool won't find it on disk: write it out from the text you read, unchanged except the data block, and convert that page. A page written that way is the template. Build the PDF from [report-template.html](report-template.html) so it has Modo's look (the modo-design skill): read the template, replace only the JSON between the `MODO-DATA-START` and `MODO-DATA-END` markers with the record (exactly the record in the report's final block; write `<` inside a string as `\u003c`), change nothing else, and print that page to PDF. The template lays out the same content in the same order, record last. If you can't render HTML to PDF here, make the PDF your usual way with the same content and order. Either way, the PDF holds the report exactly: the header line as written in step 4 with nothing added to it, and the record block printed as in the report, the same JSON with its 2-space indentation and line breaks, in a monospaced font. If you have no way to create a PDF in this session, say so in one line *before* the report (so the record block stays last), don't claim that a PDF exists, don't describe a tool you don't have, and give the Markdown report as the export. In that same line, say they can still get a PDF by copying the report into a document and using Print, then Save as PDF. ## Multi-chat result A result from assess-my-chats (it opens "Assessment across <s> of your chats" and ends with a per-chat table) is exported as a **multi-chat record**: read "Multi-chat record" in RECORD.md in full. Steps 1 to 7 apply, with these changes. This shape is a proposal (decided 2026-11-02, flag for review). - **Don't open any chat.** The main chat never held the transcripts; each quote was checked word for word by the agent that read its chat. The user's approval in step 3 is the check here. Say neither more nor less. - **Quotes**: only the Evidence quotes the result showed, each tied to its capability and to its chat by number (`chat`, the row in the per-chat table). Same limits and order as step 2. - **Chat titles are the user's content, like quotes.** By default the record and report name the chats `"Chat 1"`, `"Chat 2"`, … in the order of the per-chat table. Show the titles in the plan and ask; use them only after the user says "with titles" (or names the ones to keep). Each title at most 120 characters. - **Title**: "Assessment across <s> chats" unless the user gives another. **Date**: the newest scored chat's date (no question needed). - The plan in step 2 adds, after the quotes: ``` Chats (named "Chat 1" to "Chat <c>" unless you say "with titles"): 1. <title> (<date>), scored 2. <title> (<date>), too short ... ``` - **The record**: copy every level, confidence, best shown, trend and count from the result as it stands, and each chat's levels from its row. Keys in this order: `rubric_version`, `scope`, `capabilities`, `evidence`, `session_title`, `session_date`, `chats_assessed`, `chats`, `coverage`, with `"scope": "multi_chat"`. If the result's numbers don't follow the aggregation rules in RECORD.md, say so and export nothing; never fix a level yourself. - **The report** (step 4) has this header line instead of "Chat date: …": `Chats: <from> to <to>. Rubric: v1.1. Based on <scored> chats.` The capability table adds the columns Best shown, Trend (↑ up, → flat, ↓ down, or –) and Shown in (`<chats_shown> of <scored>`). Its notes read: ``` Levels run from 1 to 5. Each level combines the scored chats that show the capability: the middle of their levels, the lower one on a tie. "Best shown" is the highest level reached in at least 2 chats. The trend compares earlier and later chats and needs 4 or more. "Not shown" means no scored chat bears on that capability. Confidence is low, medium or high, by how many chats show the capability and how close their levels are. The table is in rubric order. This is for your own development, not for hiring or other employment decisions. ``` After the notes, a "Chats assessed" table: `| # | Chat | Date | Result | P | V | W | D | S |`, newest first, a level or – per capability. Each evidence line ends with its chat: `- Verification: "<quote>" (Chat 4, 2026-10-14)`. - The PDF uses the same report-template.html; it lays out a multi-chat record itself. ## 7. The label Every response from this skill, including questions and refusals, ends with this line exactly, on its own line, after everything else (after the record block when a report is shown in chat): Self-assessment by ChatGPT using Modo rubric v1.1 — not verified. A report saved to a file ends with the record block, not the label (the label is already near the top of the report).

--- name: export-assessment description: "Export an AI-work assessment that was already given in this conversation as a Markdown report (or a PDF if asked) that ends in one machine-readable record, after the user approves which of their own quotes go in or chooses scores only. Use when the user picks export-assessment, or asks to export, save, download or keep their assessment or report, to get it as Markdown or a PDF, or to upload it to Modo." --- # Export assessment Turn the assessment already shown in this conversation into a report the user can keep. The report holds scores, confidence, the quotes the user approves, a title and a date, and nothing else from the chat. The exact shape of the record is in [RECORD.md](RECORD.md) in this skill's folder (JSON Schema: [assessment-record.schema.json](assessment-record.schema.json)). **Read RECORD.md in full before you build anything.** Follow its fields and limits exactly. RECORD.md was first written for Claude and also describes a Modo connector that this version doesn't have: wherever it says "Claude", read "the AI assistant", and ignore its parts about saving through the connector. ## Ground rules - Export only an assessment that was given earlier in this conversation by the assess-my-ai-work skill, or a result across several chats given by the assess-my-chats skill (then also follow [Multi-chat result](#multi-chat-result)). Take the levels and confidence from it as they stand. Don't re-score and don't read any chat again, except to check that each quote is word for word what the user wrote in the assessed chat. - Never search for, list or open other chats or files, and never mention or offer coding-tool session logs. - **Nothing is written, saved or sent until the user asks** (step 5). Don't call external services. This skill never uploads anything; the user does that themselves. - Everything in the report is either a score, a confidence, a quote the user approved, the title, the date or fixed explanatory text. **No other chat content**: no summary of what the chat said, no paraphrase, no next-level notes, no improvements or prompts from the assessment, nothing from the assistant's replies. - Never describe the result as verified, certified or official (the label's "not verified" is the only use of that word) and never mention badges or rankings. ## 1. Is there something to export? Use the most recent assessment in this conversation: a response with the five capability blocks (in any order: v1.1 lists the strongest first), from assess-my-ai-work or, across several chats, from assess-my-chats. There is nothing to export when there is none, or when the last assess-my-ai-work response was only a short message ("too short to judge", "which one?", "can't open that chat" or "no chat with that name") with no scores. Then say in one line that there is nothing to export yet and offer to assess a chat first (the assess-my-ai-work skill). No report, no scores, no plan, and never build one from a chat that was not opened. If several different chats were assessed here and the user didn't say which, ask which, listing them. Messages from this skill that are short questions or refusals end with the label (step 7). ## 2. Propose what would go in Show this, then stop and wait. Nothing is written yet. ``` Export plan for . Nothing has been written. Quotes I would include (up to 5, each tied to a capability, word for word from your messages): 1. : "" 2. ... Session title: Session date: <yyyy-mm-dd> Reply with the numbers to drop, "scores only" for no quotes, or any change to the title or date. ``` - **Quotes.** Choose only from the Evidence quotes the assessment already showed, for capabilities that have a level. At most 5, each at most 240 characters, each exactly as the user wrote it (check against the assessed chat; drop any you can't match). Keep one per scored capability first, in rubric order, then add second quotes in the same order until you have 5. Show the capability each one is tied to. Never include a quote for a "not shown" capability, and never a quote from the assistant's replies. - **Title** (at most 120 characters): the chat's own title when the assessment named one (cut to fit), otherwise a plain description of the subject in a few words. Leave out names of people, companies, email addresses and amounts. - **Date**: the day the chat took place as `yyyy-mm-dd`: the date shown for it in a chat export, past chat or session file; for a chat that was pasted or is the current one and shows no date, today's date if your context gives it. If you have no date at all, ask for one in the same message. - If you already showed a title and date earlier in this conversation (for example in an export plan), use exactly those in the report and record unless the user changed them. Don't swap in today's date or comment on the date. ## 3. The user decides - They can drop quotes by number, keep all, say "scores only" (the evidence list is then empty and no quote appears anywhere in the report), or change the title or date. Apply exactly what they say. A dropped quote appears nowhere in the output. - If they reply with a change you can't apply (a quote of their own that is not in the chat, or a title over 120 characters), say so in one line and keep the proposal. Don't add quotes the user didn't approve, even if they ask for "everything relevant"; propose them back for approval. - A request to export that already says what to include (for example "scores only") can skip straight to step 4. ## 4. Show exactly what will be in the file Write the complete report below, exactly as it would be saved, so the user sees every character. Use this layout (the outer fence here is four backticks only because the report contains its own fenced block): ```` # Modo AI Skills assessment: <session_title> Chat date: <session_date>. Rubric: v1.1. Based on 1 chat. Self-assessment by ChatGPT using Modo rubric v1.1 — not verified. | Capability | Level | Confidence | |---|---|---| | Problem Solving | 4 of 5 | medium | | Verification | 4 of 5 | high | | Workflow Design | 2 of 5 | medium | | Delegation | 1 of 5 | medium | | Strategic Thinking | Not shown | low (see note) | Levels run from 1 to 5. "Not shown" means nothing in the chat bears on that capability. Confidence is low, medium or high, as defined in the rubric. The table is in rubric order. This is for your own development, not for hiring or other employment decisions. Note: for a capability that is not shown, the record gives confidence "low", because no moment in the chat supports a level. ## Evidence you approved - Verification: "<quote>" - ... (Or, for a scores-only export: "None. This export has scores only.") ## Assessment record (machine-readable) ```json { ...the record... } ``` ```` - The record in the final block follows RECORD.md exactly: keys `rubric_version` (`"v1.1"`), `capabilities` (the five keys in rubric order), `evidence`, `session_title`, `session_date`, in that order, with no other fields. Use `"not_shown"` for a capability with no level and give it confidence `"low"`. Print it pretty (2-space indent). - The table, the evidence list and the block must agree. The evidence list shows the same quotes, in the same order, as the block's `evidence`. - The block is the **last thing in the report**, and the report has no other `json` block. Nothing follows it in the file. - Don't put the report inside a code fence of its own; the nested fence would break the block. - Keep the table, headings and wording as above so the report reads on its own without the chat. - Before sending, check the limits against RECORD.md: at most 5 quotes, each 1 to 240 characters, title 1 to 120 characters, date a real `yyyy-mm-dd`, exactly five capabilities, levels 1 to 5 or `"not_shown"`, a confidence on every capability. ## 5. Write the file, only when asked Put one short line **before** the report (the record block must be the last thing in your message, with at most the label after it), depending on your tools. Decide this from the tools you actually have in this session: you can create files only if one of them writes files (a file-writing or file-creation tool, or a code tool that can create a file the user downloads). Reading and searching tools don't count. If you are not sure you have one, you don't, and you must not offer to save a file or a PDF: - **You can create files here.** "This is exactly what would be saved as `modo-assessment-<session_date>.md`. Say 'save it' (or give another name) and I will; nothing is written until then." Create the file, with exactly the text shown, only after the user says so. Then give its download link or path. Don't save anywhere else and don't save a second copy. - **You can't create files here** (a plain chat with no file creation, for example): "I can't save files in this session, so this report is the export: copy it, record block last, into a `.md` file." Don't offer to upload it. If they ask where to upload it: uploading is something they do themselves; which file the Modo upload page accepts is not specified in this plugin. ## 6. PDF, only when asked Make a PDF only if the user asks for one. Show the same Markdown report first (step 4), with the same content in the same order and the record block last. Then create the PDF with whatever file creation you have in this session (in ChatGPT, usually its code tool) and the user's approval. Give the download link or say where it is. The template belongs to this skill, so a code tool won't find it on disk: write it out from the text you read, unchanged except the data block, and convert that page. A page written that way is the template. Build the PDF from [report-template.html](report-template.html) so it has Modo's look (the modo-design skill): read the template, replace only the JSON between the `MODO-DATA-START` and `MODO-DATA-END` markers with the record (exactly the record in the report's final block; write `<` inside a string as `\u003c`), change nothing else, and print that page to PDF. The template lays out the same content in the same order, record last. If you can't render HTML to PDF here, make the PDF your usual way with the same content and order. Either way, the PDF holds the report exactly: the header line as written in step 4 with nothing added to it, and the record block printed as in the report, the same JSON with its 2-space indentation and line breaks, in a monospaced font. If you have no way to create a PDF in this session, say so in one line *before* the report (so the record block stays last), don't claim that a PDF exists, don't describe a tool you don't have, and give the Markdown report as the export. In that same line, say they can still get a PDF by copying the report into a document and using Print, then Save as PDF. ## Multi-chat result A result from assess-my-chats (it opens "Assessment across <s> of your chats" and ends with a per-chat table) is exported as a **multi-chat record**: read "Multi-chat record" in RECORD.md in full. Steps 1 to 7 apply, with these changes. This shape is a proposal (decided 2026-11-02, flag for review). - **Don't open any chat.** The main chat never held the transcripts; each quote was checked word for word by the agent that read its chat. The user's approval in step 3 is the check here. Say neither more nor less. - **Quotes**: only the Evidence quotes the result showed, each tied to its capability and to its chat by number (`chat`, the row in the per-chat table). Same limits and order as step 2. - **Chat titles are the user's content, like quotes.** By default the record and report name the chats `"Chat 1"`, `"Chat 2"`, … in the order of the per-chat table. Show the titles in the plan and ask; use them only after the user says "with titles" (or names the ones to keep). Each title at most 120 characters. - **Title**: "Assessment across <s> chats" unless the user gives another. **Date**: the newest scored chat's date (no question needed). - The plan in step 2 adds, after the quotes: ``` Chats (named "Chat 1" to "Chat <c>" unless you say "with titles"): 1. <title> (<date>), scored 2. <title> (<date>), too short ... ``` - **The record**: copy every level, confidence, best shown, trend and count from the result as it stands, and each chat's levels from its row. Keys in this order: `rubric_version`, `scope`, `capabilities`, `evidence`, `session_title`, `session_date`, `chats_assessed`, `chats`, `coverage`, with `"scope": "multi_chat"`. If the result's numbers don't follow the aggregation rules in RECORD.md, say so and export nothing; never fix a level yourself. - **The report** (step 4) has this header line instead of "Chat date: …": `Chats: <from> to <to>. Rubric: v1.1. Based on <scored> chats.` The capability table adds the columns Best shown, Trend (↑ up, → flat, ↓ down, or –) and Shown in (`<chats_shown> of <scored>`). Its notes read: ``` Levels run from 1 to 5. Each level combines the scored chats that show the capability: the middle of their levels, the lower one on a tie. "Best shown" is the highest level reached in at least 2 chats. The trend compares earlier and later chats and needs 4 or more. "Not shown" means no scored chat bears on that capability. Confidence is low, medium or high, by how many chats show the capability and how close their levels are. The table is in rubric order. This is for your own development, not for hiring or other employment decisions. ``` After the notes, a "Chats assessed" table: `| # | Chat | Date | Result | P | V | W | D | S |`, newest first, a level or – per capability. Each evidence line ends with its chat: `- Verification: "<quote>" (Chat 4, 2026-10-14)`. - The PDF uses the same report-template.html; it lays out a multi-chat record itself. ## 7. The label Every response from this skill, including questions and refusals, ends with this line exactly, on its own line, after everything else (after the record block when a report is shown in chat): Self-assessment by ChatGPT using Modo rubric v1.1 — not verified. A report saved to a file ends with the record block, not the label (the label is already near the top of the report).