| name | add-entry |
| description | Add a new resource to the Awesome Multi-Agent AI Safety list with full verification. Use when asked to add a paper, tool, benchmark, testbed, or organisation to the list. |
Add an entry to the list
Follow these steps in order. Do not skip verification.
1. Scope gate
Confirm the resource's primary subject is multi-agent AI safety (interacting agents, inter-agent attacks, agent trust infrastructure, population oversight, multi-agent testbeds). If it is single-agent safety content, stop and tell the user it is out of scope per AGENTS.md — do not add it.
For borderline cases, launch the entry-reviewer agent and follow its verdict.
2. Verify the link
- Fetch the URL (WebFetch) and confirm it resolves to the claimed resource.
- Never guess URLs or arXiv IDs. For papers, link the arXiv abstract page (
https://arxiv.org/abs/<id>), not the PDF.
- If the user supplied no URL, find the canonical one via WebSearch and verify it.
3. Check for duplicates
Grep README.md for the URL (and obvious variants) and the resource name.
4. Write the entry
- **[Name](https://link)** - One-sentence description ending with a full stop. *tag / tag*
- One sentence, technically precise, neutral, en-GB spelling.
- Tags: two, lowercase apart from proper nouns, separated by
/ (e.g. Paper / covert communication, Environment / cooperation).
- Say what the resource is and does, not why it is impressive.
5. Place it
Insert into the single best-fit section, matching the ordering convention of that section (foundational/survey works first, then specific work).
6. Update the count and validate
- Increment the "contain N entries" number in the README header.
- Run
npm run validate — it checks lint, the count, TOC anchors, and duplicate URLs. Fix anything it reports.
7. Report
Tell the user: the entry line as added, the section, the new total count, and that validation passed.
