Skill Claude CodeCodex
Plan a feature - gathers context, researches, interviews, identifies risks, outputs PRD. Use when: user says autoplan, /autoplan, or wants to plan a feature with the full research-interview-premortem workflow.
Skill Claude CodeCodex
Autosymph monitor cockpit — observe, diagnose, and heal agent loops. Attaches to a running autosymph instance via the status API, detects stalls/churn/crashes, self-heals via structify/investigation, and mutates Linear. Use when: user says /autosymph-monitor, 'monitor autosymph', 'babysit', or 'watch the agents'.
Skill Claude CodeCodex
Handle large intermediate data safely in long-running agent sessions. Prevents context compaction from silently corrupting base64, JSON, test output, and other data passed between API turns. Use when capturing screenshots, recording videos, uploading artifacts, or running long test suites.
Skill Claude CodeCodex
Record demo videos using real screen recording + screenshot stitching hybrid approach. Captures smooth video clips per action, overlays color-coded subtitle badges, and concatenates into a polished demo. Works with iOS Simulator MCP (recordvideo) and macOS screencapture (web). Triggers on "record a video", "screen…
Skill Claude CodeCodex
Upload screenshots to Linear from iOS Simulator or Playwright. Handles resize, base64 encode, upload, and embed URL generation. Use when capturing visual evidence for Linear issues during verification or implementation.
Skill Claude CodeCodex
Turn a recurring agentic failure into a deterministic skill with a test. Investigates the gap, compares working vs failing runs, writes a skill + test.sh, hardens prompts/config, and writes a postmortem. Use when an agent pipeline fails due to infrastructure, skill, or config gaps — not code bugs. Triggers on…
Skill Claude CodeCodex
Use during verifyreview (or after any verify run) to mechanically check whether the verify agent honored its completion contract — posted a summary, uploaded screenshots, no orphan local files. Returns a JSON+markdown audit and exit 0 (clean) or 2 (violations to cite as rejectstructify reasons).
Skill Claude CodeCodex
MANDATORY end-action for the verify agent — script-driven completion that uploads all visual evidence to Linear AND posts the verification summary as ONE atomic operation. Replaces hand-rolled savecomment + createattachment loops that can fail mechanically across long verify runs. Triggers on "verify finalize"…
Skill Claude CodeCodex
Use BEFORE Phase 1 of any verify run. Probes infrastructure (auth.md fixture, idb/simulator, Playwright) in <10s and fast-fails the run if anything is broken instead of stalling for 10 minutes discovering it the hard way. Triggers on "verify preflight", auto-invoked from verify prompt Phase 0.