What problem does it solve? Long unattended autonomous runs lose their mission and rules each time the agent's context is compacted and summarized. This Skill provides a reload-resilient constitution that re-grounds the agent after every compaction so a multi-wave backlog-clearing marathon never drifts, duplicates work, or violates its guardrails. ## Core Features & Use Cases - Reload contract: Re-invocation after every compaction restores the mission, granted authority, per-issue loop, and hard guardrails from a single self-contained file, with a STATE header in the session log as the only mandatory post-compaction read. - Autonomous per-issue workflow: Sync from main, triage issues, implement with TDD, open PRs, run mandatory review, and self-merge only under a strict merge gate (clean review, green CI on the final commit, all threads resolved). - Session and cost discipline: Wave-boundary context shedding, idempotent re-entry derived from GitHub state, and a per-wave cost circuit breaker as the sole sanctioned stop condition. - Use Case: You launch an unattended overnight run to clear a GitHub issue backlog; after each context compaction the agent re-invokes the directive, reads the STATE header, and safely resumes the next issue without re-reading the full history. ## Quick Start Invoke /prime-directive at the start of an unattended run and again after every compaction to reload the mission and resume the backlog marathon.
