Frc Pitcrew

Worker Plans

• build_deploy_error -> PatternScout, Librarian, Builder, Checker, Arbiter
• api_docs_lookup -> PatternScout, Librarian
• subsystem_or_command_draft -> PatternScout, Librarian, Builder, Checker, Arbiter
• autonomous_or_pathing -> PatternScout, Librarian, Builder, Checker, Arbiter
• sensor_or_can_fault -> PatternScout, Librarian, Arbiter
• vision_problem -> PatternScout, Librarian, Builder, Checker, Arbiter
• simulation_or_halsim -> PatternScout, Librarian, Builder, Checker, Arbiter
• explain_or_review -> PatternScout, Librarian (+ Arbiter for review/safety hints)
• deep_debug -> PatternScout, Librarian, DeepDebug
• follow_up -> dynamic follow-up policy
• general_or_non_frc -> direct answer, no workers

Follow-up Policy

• Attach parent dossier when likely follow-up language appears.
• If follow-up failure comes after reviewed/escalated response, route to PatternScout -> Librarian -> Arbiter -> DeepDebug.
• Otherwise recurse to parent intent plan when available.

Execution Rules

• Default worker call style: sessions_spawn.
• sessions_send reserved for deliberate persistent context.
• PatternScout and Checker may skip gracefully if unavailable.
• Arbiter must not be silently skipped on substantive/review-worthy flows.
• If Arbiter is unavailable, answer must be guarded_answer.
• Never claim review/checks that did not happen.

Show Work Mode

When user requests evidence/show-work, include:

• retrieval sources and coverage note
• review verdict and concern list
• what checks passed/failed/skipped
• explicit uncertainty if guarded