# ARSENAL: five Claude skills — source ledger

- Approved guide slug: `5-skills-claude` (not `last30days`).
- Languages: `en`, `pt-br`, `es` under `src/content/learn/5-skills-claude/`.
- Evidence checked live: **2026-10-21**. Retrieval is not an end-to-end product test.
- All installation, prompts, agent orchestration, HTML lesson/figure generation, research runs, optional integrations and paid services: **NOT TESTED BY US — Pendente: teste manual**.
- We did not log into Claude or third-party providers, access browser cookies, create accounts, or execute downloaded upstream scripts.
- The public copy of this ledger is `/learn/5-skills-claude/sources.md`.

## Official upstream repositories and pinned revisions

“Official” here means the creator's own project repository, not an Anthropic endorsement. GitHub repository metadata and branch commit APIs were retrieved anonymously; archive downloads use the immutable commit below.

| Skill | Creator repository | Retrieved commit | Included scope |
| --- | --- | --- | --- |
| Chief of Staff | https://github.com/mattpocock/skills | `5beb35b040e3de97e54fa9e386d7e42d15c295c2` | Entire `skills/in-progress/chief-of-staff/`, root README and MIT license |
| Teach | https://github.com/mattpocock/skills | `5beb35b040e3de97e54fa9e386d7e42d15c295c2` | Entire `skills/productivity/teach/`, root README and MIT license |
| Anatomy | https://github.com/wheresryan22/anatomy | `1482e8469b0a9e0d8cc18ea2740a01423d29ae80` | Entire repository's regular files, including kit, scripts, references, examples and MIT license |
| Marketing Psychology | https://github.com/coreyhaines31/marketingskills | `e85f3cbc2e4a79bf33b2a12e9bdd9087d414735b` | Entire `skills/marketing-psychology/`, root README and MIT license |
| last30days | https://github.com/mvanhorn/last30days-skill | `832d09591568e37dd990ccb9d5c51fb4a403f5ed` | Runtime complete snapshot excluding promotional media (`media/pr-assets/` omitted); nested runtime skill, engine, templates, references and MIT license retained |

Each ZIP adds a clearly labeled `CAUHI-ATTRIBUTION.txt`; upstream file bytes are unmodified. No `.git` history is bundled. Root repository READMEs may link to files outside the scope of the three single-skill downloads; the named skill directories themselves are complete. Preserve original notices when redistributing. The last30days download is a runtime complete snapshot excluding promotional media: `media/pr-assets/gogcli-589-zoom-demo.gif` and `media/pr-assets/last30days-ad.gif` are omitted. We do not rehost third-party videos, including animated promotional GIFs, regardless of open-source licensing. Retained examples may have their own attribution references; the root MIT license is not a claim of ownership over unrelated brands or linked external content. Upstream README links to omitted promotional assets may not resolve inside the archive.

## What each source supports

### Chief of Staff

- [Pinned SKILL.md](https://github.com/mattpocock/skills/blob/5beb35b040e3de97e54fa9e386d7e42d15c295c2/skills/in-progress/chief-of-staff/SKILL.md).
- Explicit `in-progress` placement is the basis for the experimental warning.
- `disable-model-invocation: true`: invoke `/chief-of-staff` explicitly in Claude Code.
- Coordinates subagents; instructs using background agents and sparse communication via context pointers.
- Wiki: agree location before first write; `index.md`, linked pages, dated claims and source references; maintenance via ingest/query/lint.
- Schedules are explicitly “harness-permitting,” not a bundled scheduler or uptime guarantee.
- No claim that it integrates email/calendar, runs after session exit or operates safely without supervision.
- **Pendente: teste manual:** discover/install experimental nested path, invoke skill, inspect actual background-agent behavior and wiki writes, confirm scheduling availability in host.

### Teach

- [Pinned SKILL.md and templates](https://github.com/mattpocock/skills/tree/5beb35b040e3de97e54fa9e386d7e42d15c295c2/skills/productivity/teach).
- Teaches across sessions using workspace files: `MISSION.md`, `RESOURCES.md`, `lessons/*.html`, `reference/*.html`, `learning-records/*.md`, shared `assets/`, `NOTES.md`.
- Must retain the supporting `*-FORMAT.md` templates; paths resolve from invocation workspace except template links.
- Sources-first teaching; short lessons, references, exercises, feedback, retrieval practice, spacing and interleaving.
- **False core claim rejected:** automatic spaced repetition/reminders. The skill describes learning principles, not an automatic scheduling system.
- `disable-model-invocation: true`: explicit `/teach`.
- **Pendente: teste manual:** install, create first lesson and cheat sheet, open HTML and assets, return in later session, inspect persistence and source quality.

### Anatomy

- [Pinned README](https://github.com/wheresryan22/anatomy/blob/1482e8469b0a9e0d8cc18ea2740a01423d29ae80/README.md), especially What it makes, Install, Requirements and Limits.
- [Pinned SKILL.md](https://github.com/wheresryan22/anatomy/blob/1482e8469b0a9e0d8cc18ea2740a01423d29ae80/SKILL.md).
- Isometric SVG, SVG+WebGL effects, true 3D moving-parts mode explicitly **beta**.
- Standalone HTML/SVG and React output; mechanism/object orientation, not a general charting library.
- Build/runtime Node.js **20.10+**; browser verification Chrome/Chromium; `CHROME_PATH` override; React requirement only for React output.
- Distinct installer minimum: current Vercel skills README requires **22.20+**; Anatomy README also calls this out. Do not tell beginners Node 20 is enough for `npx skills`.
- Upstream says long sessions; its 15–20 minute 30-part full 3D orbit audit is an **upstream claim**, not our measured runtime or a generation SLA.
- No additional API key required for the kit itself; Claude usage still costs/has limits.
- Screenshot reference `/learn/5-skills-claude/anatomy-public.webp` shows the public GitHub repository/documentation, not an examples site or generated work by us; see screenshot provenance below.
- **Pendente: teste manual:** install, SVG generation, Chrome checks, open final HTML, test WebGL and reduced-motion/mobile behavior, separately evaluate beta 3D.

### Marketing Psychology

- [Pinned SKILL.md](https://github.com/coreyhaines31/marketingskills/blob/e85f3cbc2e4a79bf33b2a12e9bdd9087d414735b/skills/marketing-psychology/SKILL.md).
- Applies psychological principles/mental models to marketing with ethical implementation guidance.
- Checks for product context in `.agents/product-marketing.md`, with documented `.claude` and legacy fallbacks.
- Count verified programmatically from exact downloaded text with regex `^### (.+)$` in multiline mode: **72 headings** on 2026-10-21. This verifies “70+” as a heading count, not scientific independence or universal empirical validity. Overlapping concepts exist (e.g. mental accounting in two contexts).
- No promise of improved conversion, analytics integration, campaign publication, or an extra API requirement for the skill itself.
- **Pendente: teste manual:** install/invoke, assess applicability, ethical copy review, create and measure a real controlled experiment.

### last30days

- [Pinned README](https://github.com/mvanhorn/last30days-skill/blob/832d09591568e37dd990ccb9d5c51fb4a403f5ed/README.md).
- [Authoritative pinned runtime SKILL.md](https://github.com/mvanhorn/last30days-skill/blob/832d09591568e37dd990ccb9d5c51fb4a403f5ed/skills/last30days/SKILL.md), version `3.28.1`.
- [Setup wizard](https://github.com/mvanhorn/last30days-skill/blob/832d09591568e37dd990ccb9d5c51fb4a403f5ed/skills/last30days/references/setup-wizard.md), [research runbook](https://github.com/mvanhorn/last30days-skill/blob/832d09591568e37dd990ccb9d5c51fb4a403f5ed/skills/last30days/references/research-runbook.md), [configuration](https://github.com/mvanhorn/last30days-skill/blob/832d09591568e37dd990ccb9d5c51fb4a403f5ed/CONFIGURATION.md), [security](https://github.com/mvanhorn/last30days-skill/blob/832d09591568e37dd990ccb9d5c51fb4a403f5ed/SECURITY.md).
- Recommended Claude Code installation is plugin marketplace add/install. Generic skills CLI is supported. We also provide a pinned ZIP route that explicitly copies the **nested** `skills/last30days` folder; there is **no root SKILL.md** in the repository snapshot. Do not install the whole repo root as a manual skill.
- Engine requires **Python 3.12+** before research/helper commands; host resolves its installed paths. No invented fixed plugin-cache paths in guide.
- No-setup starting sources documented as Reddit/HN/Polymarket/GitHub/web. Actual availability may be rate-limited, fail or depend on host web search; not a complete social firehose.
- X requires authenticated/provider path or explicit consent to supported browser sessions. YouTube public search/transcripts use `yt-dlp`, requiring installation/visible PATH; no claim YouTube is key-required by default.
- TikTok/Instagram require `SCRAPECREATORS_API_KEY`. Optional keyed Reddit and YouTube fallback/rescue can spend credits too.
- Config: `~/.config/last30days/.env`; trusted project `.claude/last30days.env`; slash-command save default `~/Documents/Last30Days`; explicit empty memory dir disables brief saves. Bare engine invocation has different save defaults; guide does not offer a raw engine one-liner.
- `BROWSER_CONSENT=false`/`FROM_BROWSER=off` disable documented direct cookie routes; SECURITY.md additionally says set `AGENTCOOKIE=off` to disable that route. All three included in guide example.
- `LAST30DAYS_REDDIT_SC_MIN_ITEMS=0` and `LAST30DAYS_YT_SC_MIN_ITEMS=0`: empty-only **search** backfill, NOT complete paid-call disabling. Transcript rescue still costs.
- `--days=7`, `--quick`, `--deep` supported by runbook; no guarantee of returned item counts.
- Source doctor is distinct from Claude Code `/doctor`; request health via the skill.
- HTML publishing is explicit opt-in and may be public/indexed. The guide defaults to local files.
- Screenshot reference `/learn/5-skills-claude/last30days-public.webp`: public GitHub repository/documentation capture, not our research output; see screenshot provenance below.
- **Pendente: teste manual:** every install route, onboarding, skipped-cookie behavior, actual source health/coverage, all live research, optional API integrations, exports, billing/credit consumption.

## Important live-source discrepancy: ScrapeCreators free quota

Checked 2026-10-21:

- last30days README/runtime/setup wizard advertise **10,000 free calls**, then pay as you go.
- The provider's own current public homepage, https://scrapecreators.com/, says **100 free starting credits**, and “Claim up to 7,000 more” through conditional rewards (how you found them, GitHub star, feedback/review).
- Verified in both homepage HTML visible copy and its FAQ JSON-LD, retrieved directly with Python urllib. The JSON-LD answer says: “you get 100 free credits with no credit card required.”
- The dedicated `/pricing` URL could not be extracted; homepage pricing section was fetched directly as the alternative. We did not invent an unavailable pricing page response.
- We did not authenticate or perform device-code authorization; a skill-partner entitlement could differ. The guide explicitly does **not** promise 10,000 calls and directs readers to their account dashboard before optional setup.
- Credits and calls are not interchangeable for every endpoint. Do not label the service unlimited/free forever.
- **Pendente: teste manual:** logged-in account quota, any skill-specific partner offer, endpoint credit usage.

## Claude Code prerequisites and pricing

Checked 2026-10-21:

- https://code.claude.com/docs/en/quickstart — native installer, sign-in, supported subscriptions/Console account/cloud access; desktop/web/IDE interfaces exist but are not claimed tested here.
- https://code.claude.com/docs/en/skills — `.claude/skills/<name>/SKILL.md` per project; `~/.claude/skills/` personal; commands `/skill-name`, supporting files and invocation controls. Skill features can differ across hosts.
- https://claude.com/pricing — Pro **US$20 monthly**, or **US$200/year prepaid** (page displays US$17/month); includes Claude Code with usage limits; ordinary Free plan comparison says no Claude Code. Max starts **US$100/month**. Taxes excluded, prices/plans may change. No fabricated BRL conversion.
- Claude subscription is not an OpenAI/xAI/ScrapeCreators API key. No statement that any subscription is unlimited.
- **Pendente: teste manual:** install/login, subscription entitlement in checkout/region, usage limits, mobile/remote compatibility for these skills.

## Installer source

- https://github.com/vercel-labs/skills/blob/main/README.md and https://github.com/vercel-labs/skills/blob/main/docs/commands.md, fetched live 2026-10-21.
- Current README: Node.js **22.20+**.
- Documented source formats include full GitHub tree URL and local path; options include `-a claude-code`, `--skill`, project default and `-g` user-wide scope.
- Guide uses direct tree URLs for deeply nested Chief of Staff and Teach rather than relying on recursive repository discovery.
- `list`, `check`, `update`, `remove` documented; no bypass-confirmation defaults in guide.
- **Pendente: teste manual:** all published install commands. Our source/archive verification is not CLI installation verification.

## Screenshot provenance

Capture evidence supplied by the parent agent: both public GitHub pages returned **HTTP 200** on **21 October 2026**, captured with **Playwright using Google Chrome** at a **1440 × 1000** viewport and saved as **WebP**. These are public repository/documentation references, not installed-skill tests, generated results, or rehosted promotional videos.

| Public image | Captured source URL |
| --- | --- |
| `/learn/5-skills-claude/anatomy-public.webp` | https://github.com/wheresryan22/anatomy |
| `/learn/5-skills-claude/last30days-public.webp` | https://github.com/mvanhorn/last30days-skill |

## Download inventory and integrity

Each archive was opened locally and checked with ZIP CRC validation. SHA-256 hashes cover the actual ZIP bytes. `manifest.json` supplies machine-readable metadata. Counts include the added attribution file, root notices and README where specified.

| Public URL | Files | Bytes | SHA-256 |
| --- | --- | --- | --- |
| `/learn/5-skills-claude/chief-of-staff.zip` | 5 | 10238 | `7261201a2a3e2dd802af3220249edeb1dd3ce525cbbfe3eec46fcc07d2e93809` |
| `/learn/5-skills-claude/teach.zip` | 9 | 17203 | `e12936c39bd03f52845f8bb289fb36fd6ad262b468f300bf8f71b22b3459e14c` |
| `/learn/5-skills-claude/anatomy.zip` | 174 | 14278620 | `e0e059c4824196afef24ee602277357cff8a2d09e9f5cd8edd553b96bd62e702` |
| `/learn/5-skills-claude/marketing-psychology.zip` | 5 | 20639 | `1d21bf35040d25230a358d01dabe045d22cea5c1b5074e57c6f84f0f93fa68fa` |
| `/learn/5-skills-claude/last30days.zip` | 208 | 1191403 | `8c897312012416abd886a0e81e70f8ce096bec0e2a382eee1f157c05c28a9e7d` |

## Editorial boundaries

- All three guides contain installation instructions for **each** skill, one bounded prompt, expected artifacts framed as untested, local downloads, separate advanced configuration/commands, and FAQ covering cost, computer/mobile, safety, troubleshooting and limitations.
- No invented console output, generated-result screenshots, personal trial claims, star counts or biographical filler.
- On-site downloads are the primary action; GitHub URLs in commands/source references identify upstream, not a requirement to leave the guide to obtain files.
- Parent owns shared loaders/resources/curation/config/tests, screenshot capture, PDFs, app build, integration, commit and deploy. No shared code was changed by this authoring task.
- **Pendente: teste manual** remains explicit even if MDX compilation, file integrity checks or the website build pass; none substitutes for running a skill.

## Website integration verification

- Beginner bootstrap commands were independently checked against the live Anthropic quickstart on 2026-10-21: native macOS/Linux/WSL installer, Windows PowerShell installer, version check and `claude` launch. These commands were documented, not executed by us.
- Local Google Chrome smoke checks rendered all three locales at a 390 × 844 viewport with the development-only `LEARN_IGNORE_EMBARGO=1` override. Each returned HTTP 200, both public screenshots loaded, all five skill names were present, and no browser page errors were emitted. This checks the website, not the skills.
- Production build and publication checks use the embargo without that local override. No authenticated Claude session or skill execution was tested.