microsoft/cat-agent-skills is a static website that catalogs reusable instruction sets and related packages for AI agents. People use it to search, filter, rate, and download skills for Cowork, Copilot Studio, and Scout, along with Copilot plugins and Scout automations. The catalogue entries are the skills, instructions, plugins, and settings displayed by the site.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add microsoft/cat-agent-skills --skill template-based-document-generationgit clone --depth 1 https://github.com/microsoft/cat-agent-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/microsoft/cat-agent-skills/template-based-document-generation)<a href="https://agentmods.dev/skills/microsoft/cat-agent-skills/template-based-document-generation"><img src="https://agentmods.dev/badge/skills/microsoft/cat-agent-skills/template-based-document-generation/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/microsoft/cat-agent-skills/template-based-document-generation"><img src="https://agentmods.dev/badge/skills/microsoft/cat-agent-skills/template-based-document-generation.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00078 | $0.03419 |
| Opus 5.5 | $0.00031 | $0.01368 |
| Sonnet 5.5 | $0.00016 | $0.00684 |
| Haiku 4.5 | $0.00008 | $0.00342 |
Grade A, and why
template-based-document-generation scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 191 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Template-Based Document Generation
One command inventories the template, you write a plan as data, one command applies and verifies it.
Branding survives: every part the plan does not touch (logo, styles, theme, fonts, numbering, headers and footers with nothing to fill) stays byte-identical, and in a touched part only the runs holding a replaced value are rewritten. Nothing is invented silently: a missing value follows the field's absence policy and is named in the report rather than guessed, every literal, sentinel, empty and unused input is a review item, and every change is verified against the plan or the run fails and nothing is published.
Workflow
python scripts/run_job.py inspect --template <file.docx> --workdir <dir> [--profile <profile.json>]creates<dir>/<job_id>/withsnapshot.docx,manifest.json(paragraph inventory, candidates, findings) andplan.draft.json, and prints a listing: every paragraph with its id and text, every candidate marked!(needs your decision),?(optional) orx(cannot be filled, reason given).- Copy
plan.draft.jsontoplan.jsonand complete it. A complete, working pair to copy the shape from is references/examples/plan.json and references/examples/values.json: every decision kind, typed and formatted fields, a sentinel, a derivedliteral, a repeated row, a range deletion and a multi-paragraph insert. It is written forassets/fixtures/mixed-conventions.docxand selftest runs it, so every key name in it is current. Read it instead of the contract or the scripts when you need a setting's name.decisions[]: one per!candidate.field(fill it),literal(ordinary text, leave it),remove_text(delete just that text),remove_paragraph, orkeep(leave it and flag it).fields[]:name,type(string, integer, decimal, date, boolean),format,absence_policy(required, sentinel, empty) andsource, a path into the data such asclient.nameoritems[0].qty.sourceis a plain path, traversed key by key: there are no expressions, so it cannot join, compute or summarise. When the value needs working out rather than looking up, put the worked-out value in the field'sliteralinstead of asource; it is always flagged for review. (Careful:literalindecisions[]means the opposite of this one — there it means "this text is real content, leave it alone". Infields[]it is the value you supply.)operations[]:delete_paragraph,delete_range(from and to paragraph ids in the same container; tables in between go too),delete_row,delete_break(one line or page break, so deleting a cover page does not leave a blank first page),repeat_row(anid, any paragraph in the row, asourcelist; the row's fields carry"repeat": "<id>"and sources relative to each item; add"group": trueto a field such as a phase or category to print it only on the first of consecutive rows that share its value),insert_paragraph_after(textorfield; a field whose value is a list becomes one paragraph per item, which is how a multi-paragraph answer goes in at an anchor).localewhen the document is not English: months, true and false text, separators, date pattern.notes: optional free text for the record, at most 2000 characters. The draft carries a reminder here; remove or replace it.
- Write
values.jsonas{"data": <the input, any JSON, shaped however you like>}. python scripts/run_job.py execute --job <job dir> --dry-runprints every change as before and after text plus every review item, or the errors that stop it. Fix the plan and repeat until it compiles.python scripts/run_job.py execute --job <job dir> [--out <dir>]applies, verifies, and writesreport.json,attempts/<id>/and, when verification passes,output/<stem>__<job_id>.docx. Exit 0 ok, 2 needs_review, 1 failed. Rerunnable, and a failed rerun never deletes a document an earlier attempt verified.- Deliver the document before you report on it.
output/<stem>__<job_id>.docxsits in the job directory, which the user never sees: copy it into the folder this session hands deliverables from, or run step 5 with--outpointing there. A report without the file is not a delivery. If execute exits 1 there is nothing to deliver; say so and name the check that failed. - Relay the result word, the change list and every review item. Each command prints its own
elapsed:line; pass those figures on unchanged rather than estimating, if anyone is measuring.okmeans mechanically verified against the template and the plan. Nobody has opened the document in Word.
What ships with it
51 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- assets/expected/archive-debris.json 1.8 KB
- assets/expected/block-sdt-wrapping.json 2.3 KB
- assets/expected/comments-only.json 1.6 KB
- assets/expected/complex-field-result.json 1.8 KB
- assets/expected/deletion-safety.json 6.4 KB
- assets/expected/grouped-rows.json 3.4 KB
- assets/expected/macro-default-type.json 43 B
- assets/expected/mc-ignorable-root.json 2.0 KB
- assets/expected/mixed-conventions.json 15 KB
- assets/expected/mixed-format-split.json 1.5 KB
- assets/expected/nested-controls.json 2.3 KB
- assets/expected/repeated-text-off.json 1.4 KB
- assets/expected/revisions-only.json 2.0 KB
- assets/expected/signature-marker.json 636 B
- assets/expected/six-position.json 26 KB
- assets/expected/tab-in-placeholder.json 1.7 KB
- assets/expected/wrapped-control.json 3.4 KB
- assets/fixtures/archive-debris.docx 4.3 KB
- assets/fixtures/block-sdt-wrapping.docx 4.2 KB
- assets/fixtures/comments-only.docx 4.6 KB
- assets/fixtures/complex-field-result.docx 4.2 KB
- assets/fixtures/deletion-safety.docx 4.3 KB
- assets/fixtures/grouped-rows.docx 4.3 KB
- assets/fixtures/macro-default-type.docx 4.4 KB
- assets/fixtures/mc-ignorable-root.docx 4.4 KB
- assets/fixtures/mixed-conventions.docx 5.1 KB
- assets/fixtures/mixed-format-split.docx 4.2 KB
- assets/fixtures/nested-controls.docx 4.4 KB
- assets/fixtures/repeated-text-off.docx 4.2 KB
- assets/fixtures/revisions-only.docx 4.3 KB
- assets/fixtures/signature-marker.docx 4.3 KB
- assets/fixtures/six-position.docx 4.5 KB
- assets/fixtures/tab-in-placeholder.docx 4.1 KB
- assets/fixtures/wrapped-control.docx 4.3 KB
- metadata.json 488 B
- README.md 5.7 KB
- references/contract.md 20 KB
- references/examples/plan.json 3.0 KB
- references/examples/values.json 489 B
- references/policy.json 1.9 KB
- references/profile.default.json 4.2 KB
- references/template-handling.md 5.1 KB
- references/verification.md 9.1 KB
- scripts/apply_plan.py 12 KB runs code
- scripts/build_fixtures.py 32 KB runs code
- scripts/inspect_template.py 39 KB runs code
- scripts/models.py 74 KB runs code
- scripts/namespace_validation.py 7.0 KB runs code
- scripts/ooxml_common.py 47 KB runs code
- scripts/run_job.py 42 KB runs code
- scripts/verify_document.py 41 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago First seen · 191 lines · 78 tokens per session scan A 4c74c793c6e4
template-based-document-generation is a skill published in the GitHub repository microsoft/cat-agent-skills (93 stars, last pushed today), licensed MIT. It adds 78 tokens to every session and 3,419 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-20.
Other skills, from other repositories
orbit-notion
Open Orbit briefing skill — selected by the Orbit pipeline when Notion is the user's only connected connector, or when the user explicitly scopes their daily digest to Notion. Pulls the past 24 hours of document edits, comments, mentions, and database row changes from the user's authenticated Notion connection and…
instrument-data-to-allotrope
Convert laboratory instrument output files (PDF, CSV, Excel, TXT) to Allotrope Simple Model (ASM) JSON format or flattened 2D CSV. Use this skill when scientists need to standardize instrument data for LIMS systems, data lakes, or downstream analysis. Supports auto-detection of instrument types. Outputs include full…
apify-youtube-transcript-scraper
Get YouTube captions, timestamped quotations, SRT/VTT and translated transcripts for videos, channels and playlists through the harshmaur/youtube-transcript-scraper Apify Actor.
legal-hold
Issue, refresh, release, or report on legal holds — drafts the hold notice as .docx, updates legalhold fields in log.yaml, and calendars the next refresh. Use when the user says "issue a hold", "refresh hold", "release hold", or asks for a portfolio-wide hold status report.
baoyu-youtube-transcript
A tool for downloading the written captions, subtitles, chapter information, speaker labels, and cover image from a YouTube video using its URL or ID.
feishu
Work with Feishu or Lark bots, docs, sheets, bitables, approval flows, and OpenAPI/MCP setup without hardcoding credentials.