docs(#17): add skill resolution section to AGENTS.md

[fullsend-ai-coder]

Document the multi-source skill resolution chain (repo → org → platform) so agents do not flag missing local skill directories as bugs when the skill exists at org or platform level.

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

---

Closes #17

Post-script verification

• Branch is not main/master (issue-17-skill-resolution-docs)
• Secret scan passed (gitleaks — bc7beaa9a57aed031c6ef1d9755c1f2cfca516da..HEAD)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 6:19 PM UTC · Completed 6:27 PM UTC
Commit: c62e78f · View workflow run →

[fullsend-ai-retro]

PR #19 was the final step in a false-positive cascade originating on PR #12, where the review agent flagged customer-research as a missing skill dependency 4 times across 4 iterations. The human dismissed it, but the retro agent filed issue #13 as a bug. Triage confirmed it, code agent produced an incorrect fix (PR #16), and ralphbean had to intervene — explaining the skill exists at org level. The retro agent's second pass correctly diagnosed the root cause, filing issue #17. The code agent implemented it cleanly as PR #19 (AGENTS.md with skill resolution docs), merged in 5 minutes. Two upstream issues (#2955, #2956) already track the review and retro agent behavioral fixes. Two repo-level improvements remain: inline annotation of extension point skills, and review finding dedup across iterations.

Proposals filed

• Annotate extension point skills with inline YAML comments in agent frontmatter (in fullsend-ai/agents)
• Review agent should deduplicate findings across re-review iterations on the same PR (in fullsend-ai/fullsend)