docs(agents): survey Anthropic's public repositories (closes #123) - #219
Conversation
Lab members keep asking whether Claude Code is open source and which of the ~100 repositories under the anthropics org are worth their time. The answer is not obvious from GitHub: the most-starred repository (claude-code) holds no source, only an issue tracker, a changelog, bundled plugins, and example configs, while the action that runs our PR reviews (claude-code-action) is fully readable MIT code. Add chapters/ai-tools/anthropic-repos.qmd, a survey of the org as of 2026-09-09: a table of the lab-relevant repositories (purpose, language, stars, last push, detected license, verdict), what the public claude-code repository actually contains, how claude-code-action picks between agent mode and tag mode, notes on the skills, plugin marketplace, SDK, and sandbox-runtime repositories, and a closing verdict on which to adopt, read, track, trial, or skip. Harness internals are left to #122. Include it in coding-agents.qmd directly after the config-install section, append the cited repositories to references.bib, and add Codespaces and Jupyter to inst/WORDLIST. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
An adversarial review of the anthropics survey found the prose contradicting the table and two claims resting on nothing a reader could check. The claude-code section called it the most-starred coding repository while the table shows skills ahead of it, so it now says second to skills. The license column defined only one of the three values it uses, so the intro now distinguishes a proprietary notice, no license file, and a file the API could not classify (NOASSERTION, which is what the API returns for courses). The 2.1.266 changelog example now names the variable the entry names, cites CHANGELOG.md, and carries a read date, since the entry was verified against the live file. The open-issue count carries a measured stamp, and the "worth knowing about" filler is gone. Refs #123. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A second adversarial pass (agy) over the anthropics survey found three table rows that no prose group accounted for, a TypeScript SDK described both as a tracker and as a package that bundles the CLI, and several claims stated without the evidence behind them. The five groups are now six, with an infrastructure group for sandbox-runtime, devcontainer-features, and the monitoring guide; the SDK sentence says which repository bundles the CLI and which is only a tracker; the fork and archive share is a measured count (38 of 108) instead of "about a third", and the organization count is refreshed to the same reading; the npm deprecation cites the README that states it; the dxt mention now says where that repository went (modelcontextprotocol/mcpb); the skills license cell reports the single identifier the API returns, with the source-available split explained in prose; and the two raw HTML anchors to the firewall section are Quarto crossrefs, which required giving that heading in coding-agents.qmd an id. The rhetorical opener and two "worth" fillers are replaced with direct statements. Refs #123. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
Pushed with the pre-push guard's Posted by Claude Code (AI agent) --- not written by a human. |
There was a problem hiding this comment.
🟡 Changes recommended
New references.bib entries use non-standard howpublished values that conflict with the established Software/Documentation convention in the existing bibliography.
Once you've addressed the issues Copilot identified, you can request another Copilot review.
Pull request overview
This PR adds a new documentation section to the “Coding agents” chapter that surveys Anthropic’s public GitHub repositories (with emphasis on claude-code vs its closed-source CLI, and claude-code-action usage/modes), and wires it into the site alongside supporting citations and spelling allowances.
Changes:
- Add a new
ai-toolspage (anthropic-repos.qmd) summarizing whichanthropics/*repos matter to the lab and why. - Transclude the new page from
chapters/coding-agents.qmdand add a stable section id for the firewall section referenced by the new content. - Add new bibliography entries and spellcheck wordlist terms to support the new page.
File summaries
| File | Description |
|---|---|
chapters/ai-tools/anthropic-repos.qmd |
New survey page describing key Anthropic repos and how they relate to lab workflows. |
chapters/coding-agents.qmd |
Adds a new “Anthropic’s Public Repositories” section and assigns an explicit id to the firewall section for cross-references. |
references.bib |
Adds citations for Anthropic repos and specific repository documentation files referenced by the new page. |
inst/WORDLIST |
Adds “Codespaces” and “Jupyter” to satisfy spellcheck for the new content. |
Review details
Suppressed comments (6)
references.bib:321
- These repository citations use
howpublished = {GitHub repository}but elsewhere inreferences.bibGitHub repositories are tagged ashowpublished = {Software}. Using the established value improves consistency for bibliography formatting.
@misc{anthropics_skills_repo,
title = {anthropics/skills},
author = {{Anthropic}},
year = {2026},
howpublished = {GitHub repository},
url = {https://github.com/anthropics/skills},
references.bib:357
- These citations are GitHub repositories, so
howpublished = {Software}is more consistent with existingreferences.bibentries thanGitHub repository.
@misc{anthropics_claude_agent_sdk_python,
title = {anthropics/claude-agent-sdk-python},
author = {{Anthropic}},
year = {2026},
howpublished = {GitHub repository},
url = {https://github.com/anthropics/claude-agent-sdk-python},
references.bib:339
- These GitHub repository entries should use
howpublished = {Software}to match the existing pattern inreferences.bib(e.g., references.bib:105-112).
@misc{anthropics_claude_plugins_community,
title = {anthropics/claude-plugins-community},
author = {{Anthropic}},
year = {2026},
howpublished = {GitHub repository},
url = {https://github.com/anthropics/claude-plugins-community},
references.bib:375
- These GitHub repository citations should use
howpublished = {Software}to match the established convention inreferences.bib.
@misc{anthropics_anthropic_sdk_python,
title = {anthropics/anthropic-sdk-python},
author = {{Anthropic}},
year = {2026},
howpublished = {GitHub repository},
url = {https://github.com/anthropics/anthropic-sdk-python},
references.bib:293
- These
claude-code-actioncitations use non-standardhowpublishedvalues. Elsewhere inreferences.bib, GitHub repos arehowpublished = {Software}and docs pages arehowpublished = {Documentation}(e.g., references.bib:96-112).
@misc{anthropics_claude_code_action_repo,
title = {anthropics/claude-code-action},
author = {{Anthropic}},
year = {2026},
howpublished = {GitHub repository},
references.bib:312
- This repository citation should use
howpublished = {Software}to match other GitHub repository entries inreferences.bib(e.g., references.bib:105-112).
@misc{anthropics_claude_code_security_review,
title = {anthropics/claude-code-security-review},
author = {{Anthropic}},
year = {2026},
howpublished = {GitHub repository},
url = {https://github.com/anthropics/claude-code-security-review},
- Files reviewed: 4/4 changed files
- Comments generated: 2
- Review effort level: Lite
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
…nt; normalize bib howpublished The review found the license column contradicting the chapter's own methodology: `skills` has no root license file (the API reports null), so it belongs in the `none` bucket rather than Apache-2.0, and `claude-ai-mcp` has a proprietary LICENSE the API reports as NOASSERTION, which the chapter defines as `not detected`. The `examples/` bullet said eleven workflows beyond the three review variants while naming five; the directory holds eleven files in total, so the bullet now counts them once and names all eight non-review ones. Copilot flagged the new bibliography entries for one-off `howpublished` values; they now use the repo's `Software` and `Documentation` conventions. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
🟢 Approval recommended
The documentation additions are consistent with existing cross-references and bibliography conventions, and no correctness or build-blocking issues were found in the changed files.
Review details
- Files reviewed: 4/4 changed files
- Comments generated: 0 new
- Review effort level: Lite
…os-123 # Conflicts: # references.bib
There was a problem hiding this comment.
🟢 Approval recommended
The new documentation is well-integrated, references resolve to existing section IDs, and supporting bibliography/wordlist updates align with existing repo conventions.
Review details
- Files reviewed: 4/4 changed files
- Comments generated: 0 new
- Review effort level: Lite
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
The union resolver used for the sync dropped the blank line between every pre-existing entry and truncated one entry, which made pandoc reject the file and failed the build. The file is now rebuilt from the two parent versions: main's entries in main's order and formatting, then this branch's new entries appended, validated with pandoc. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
There was a problem hiding this comment.
🟢 Approval recommended
Documentation-only changes appear consistent with existing site conventions (anchors, measured timestamps, citations) and no correctness or rendering issues were identified in the modified regions.
Review details
- Files reviewed: 4/4 changed files
- Comments generated: 0 new
- Review effort level: Lite
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
…os-123 # Conflicts: # inst/WORDLIST # references.bib
|
Claude finished review — View run
Review of PR #219 (round 4): Survey
|
There was a problem hiding this comment.
🟡 Changes recommended
references.bib introduces duplicate bibliography entries for existing Anthropic plugin marketplace sources, which should be de-duplicated to keep citations consistent.
Once you've addressed the issues Copilot identified, you can request another Copilot review.
Review details
Suppressed comments (1)
references.bib:810
- This entry duplicates an existing bibliography item for the same URL (see
@misc{claude_plugins_community}at references.bib:668-675). To avoid duplicate sources and inconsistent citation keys, reuse the existing entry (optionally enriching itsnote) and delete this duplicate, then update citations inchapters/ai-tools/anthropic-repos.qmd.
howpublished = {Software},
url = {https://github.com/anthropics/claude-plugins-community},
note = {Apache License 2.0. Accessed 2026-09-09}
}
- Files reviewed: 4/4 changed files
- Comments generated: 1
- Review effort level: Lite
| howpublished = {Software}, | ||
| url = {https://github.com/anthropics/claude-plugins-official}, | ||
| note = {Apache License 2.0. Accessed 2026-09-09} | ||
| } | ||
|
|
|
Survey the
anthropicsorganization's public repositories so lab members know which ones matter to our workflow --- in particularclaude-code(what the public repo actually contains versus the closed source) andclaude-code-action(its modes and how the lab uses it).Adds
chapters/ai-tools/anthropic-repos.qmd, transcluded fromchapters/coding-agents.qmdafter the Claude Code config/install section, with a table of relevant repos and a closing "Useful to us?" section.Closes #123
🤖 Generated with Claude Code