fix(#13): remove non-existent customer-research skill

[fullsend-ai-coder]

The prioritize agent's frontmatter declared a dependency on the customer-research skill, but no skills/customer-research/ directory exists. The prompt body already uses conditional language ("if available"), so removing the frontmatter reference is sufficient.

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

---

Closes #13

Post-script verification

• Branch is not main/master (fix/remove-nonexistent-customer-research-skill)
• 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 5:54 PM UTC · Completed 6:02 PM UTC
Commit: 9a033c6 · View workflow run →

[fullsend-ai-retro]

PR #16 was a false-positive fix created by the automated pipeline (retro → triage → code agents) that attempted to remove an intentional cross-repo skill reference. The prioritize agent declares customer-research in its frontmatter, which exists as an org-level customized skill in fullsend-ai/.fullsend/customized/skills/customer-research/SKILL.md. The review agent flagged the missing local directory across 6 iterations on PR #12, the human author dismissed it, the retro agent filed issue #13 anyway, triage marked it ready-to-code, and the code agent created PR #16 to delete the reference. ralphbean closed both as incorrect — the reference is intentional. The cascade wasted tokens across 4 agent stages and required human intervention to stop. Root cause: no agent in the pipeline understood that skills can be resolved from org-level .fullsend customization, not just the local repo. Three proposals target this gap at different layers.

Proposals filed

• Document org-level skill resolution in AGENTS.md to prevent false-positive filing (in fullsend-ai/agents)
• Review agent should check org-level skill customization before flagging missing skill directories (in fullsend-ai/fullsend)
• Retro agent should not file issues for review findings the PR author explicitly dismissed (in fullsend-ai/fullsend)