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 deliverable-ingestgit 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/deliverable-ingest)<a href="https://agentmods.dev/skills/microsoft/cat-agent-skills/deliverable-ingest"><img src="https://agentmods.dev/badge/skills/microsoft/cat-agent-skills/deliverable-ingest/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/deliverable-ingest"><img src="https://agentmods.dev/badge/skills/microsoft/cat-agent-skills/deliverable-ingest.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 | $0.00175 | $0.02247 |
| Opus 5.5 | $0.00070 | $0.00899 |
| Sonnet 5.5 | $0.00035 | $0.00449 |
| Haiku 4.5 | $0.00017 | $0.00225 |
Grade A, and why
deliverable-ingest 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 yesterday.
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 — 72 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Deliverable Ingest (Connect)
Purpose
Normalize a draft deliverable and its support pack into ps.deliverable-review-quality-gate.v1: deliverable metadata, explicit recipients, located assertions, evidence items, prior findings, provenance and citations.
When to use
Start every Deliverable Review & Quality Gate flow here, especially for requests to review a deliverable, check the evidence pack, find unsupported claims or prepare a quality-gate review.
Inputs
- Draft deliverable from an engagement workspace or SharePoint.
- Supporting evidence index: models, extracts, minutes, signed client data, workshop notes, prior analysis and third-party content records.
- Intended recipient list, disclaimer block and prior review notes when supplied.
Steps
- Retrieve inputs read-only; never modify source files.
- Capture recipients exactly. If recipients are absent or ambiguous, record that state for
gate-check; do not guess names, addresses or allowed users. - Extract candidate assertions with section, page, paragraph and label, quoting the assertion text from its paragraph. Take
pagefrom the source's page breaks, or from[Page N]markers in a text extract. Number paragraphs from 1 on each page: a paragraph is a block of body text separated by a blank line, and headings, page markers and blank lines are not paragraphs. If a source is unpaginated, usepage: 1for every location, number paragraphs from 1 from the top of the document (still skipping headings), preserve the section and label, and include a missing-source note that page numbers were not present.pageandparagraphare integer physical positions counted from 1 (0 and negative numbers are input errors); keep printed labels such as roman-numeral front-matter pages (iv) inlabel. - Classify assertions as factual, quantified, forward-looking or recommendation, and set
scope_aligned(true/false, required on every assertion) from the signed SOW/evidence boundary; setsample_sizeon every recommendation (null when not sample-based). Takesample_sizeandpopulation_sizeonly from figures stated in the assertion's paragraph or its linked evidence entries; leave them null otherwise. Type every evidence item with one of the contract's evidencetypevalues; do not decide support or severity. - Record
deliverable_id,title,version,type,confidentiality_marking,client_name(and anyclient_aliases) from the deliverable cover so recipients can be positively matched to the client. Takeintended_recipientsexactly from the distribution list or named recipients; never complete or rename them. Setthird_party_recipient_named(required; true whenever any recipient is outside the client entity), thenrequired_disclaimer_type(reliance_restrictionwhen a third party is named, otherwisegeneral_advice). Takepresent_disclaimer_typesandpresent_languageonly from disclaimer and limitation text actually in the draft, such as aDisclaimer - general adviceclause. Setclient_data_handlingfrom the client-data register, usingunknownwhen no register is supplied or it cannot confirm handling. Loadbrand_checksandlicensed_contentfrom their registers; when a register is not supplied, leave the field absent (never[]). Use[]only when the register explicitly states it has no entries (for exampleNo licensed or third-party content.); a placeholder or prose line such asTBDis not a register entry and never an empty register, so treat that register as not supplied (field absent) and add a missing-source note.criteria-retrieveverifies these fields; it does not own the first extraction. - Seed
criteriabeforecriteria-retrieveruns:criteria_set_idcomes from the criteria extract header (for exampleCriteria set ID: PS-QRM-DG-2026-demo), andcriteria_sectionsstarts as the shipped floor["quality-gate-criteria.md #1-#8", "severity-rules.md #1-#4"].criteria-retrievecompares the floor with the engagement extract and may add stricter sections, but never removes floor sections. - Preserve every source as provenance in the exact form
connector:<slug>orskill:<skill-name>: a lower-case prefix, a colon and a lower-case slug of letters, digits,-,_or.with no spaces, for exampleconnector:sharepoint,connector:engagement-workspaceorskill:deliverable-ingest(evidence extracted by this skill). Evidence without this provenance cannot support any assertion (quality-gate-criteria.md #1.5). Every assertion and the deliverable also carrysourcein this form (normallyskill:deliverable-ingest), anddeliverable.deliverable_idis required. Every location needs a non-blanklabel, and aprovenanceobject, if you write one, needsengines(an empty array before any engine runs) and an ISO-8601generated_at(quality-gate-criteria.md #8.14). When a source record (an evidence item, prior finding, brand check, licensed-content entry, assertion or reviewer finding) names the engagement, deliverable or client it belongs to, carry that as the entry'sengagement_id,deliverable_idorclientexactly as written; never drop or rewrite it to match this review. An entry of another engagement, deliverable or client is held and never used, and the engines show none of its contents, only its position; another subject named anywhere else in the payload (criteria, provenance, other deliverable or top-level members) is refused (quality-gate-criteria.md #8.15). The engines refuse a payload that breaks the contract. Pass the full payload tocriteria-retrieve, thenassertion-map.
What ships with it
3 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.
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.
- yesterday First seen · 72 lines · 175 tokens per session scan A 85f795c5a29f
deliverable-ingest is a skill published in the GitHub repository microsoft/cat-agent-skills (101 stars, last pushed yesterday), licensed MIT. It adds 175 tokens to every session and 2,247 once invoked, about $0.0007 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 2026-11-07.
deliverable-ingest compared with skills from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| deliverable-ingestmicrosoft/cat-agent-skills | Skill | 101 | yesterday | MIT | A | 175 | Claude Code, Codex |
| orbit-notionnexu-io/open-design | Skill | 107k | today | Apache-2.0 | A | 117 | Claude Code, Codex |
| instrument-data-to-allotropeanthropics/knowledge-work-plugins | Skill | 35k | yesterday | Apache-2.0 | A | 123 | Claude Code |
| apify-youtube-transcript-scrapersickn33/agentic-awesome-skills | Skill | 49k | 2d ago | MIT | A | 48 | Claude Code |
| legal-holdanthropics/claude-for-legal | Skill | 9.9k | yesterday | Apache-2.0 | A | 68 | Claude Code |
| report-to-pdfasgeirtj/system_prompts_leaks | Skill | 71k | today | CC0-1.0 | A | 17 | Claude Code, Codex |
| artifact-template-minimal-letterheadasgeirtj/system_prompts_leaks | Skill | 71k | today | CC0-1.0 | A | 54 | Claude Code, Codex |