tycenjmccann
@tycenjmccann
Public Skills
run-analysis
by tycenjmccann
Full ANALYZE-mode playbook — post-run workflow analysis. The assessment rubric (planning, execution, human-in-the-loop, rework, deliverables, trend), the exact analysis.json schema save_analysis.py requires, and the knowledge-file curation step. Load on every ANALYZE invocation, and for CHAT deep-dives into how a run performed.
watch-triage
by tycenjmccann
Full WATCH-mode playbook for a stuck workflow run — stall diagnosis patterns (missed unblocks, orphan tickets never dispatched, silent/dead agents), the "did the work actually ship?" test, and the intervention decision order. Load on EVERY WATCH invocation before intervening, and for CHAT requests to fix, unstick, dispatch, or close a run.
si-synthesis
by tycenjmccann
SYNTHESIZE-mode playbook — batch pending run analyses into ONE system-improvement PRD under the SI banner. Separates agent-level gaps (fleet repo) from harness/system gaps (hub repo), writes the PRD to the fleet-imp-agent/prd/ prefix that prd-submitter watches, and marks the analyses as batched. Load on every SYNTHESIZE invocation.
crash-rca
by tycenjmccann
Root-cause analysis for crashed/dead pipeline agent runs. Use when a WATCH-mode stuck run shows dead agent sessions (agent.started with no completion, multiple session ids on one ticket, or a retry that was itself preceded by silent deaths). Produces a formal RCA and files a bug that auto-fires the bug-fix pipeline when the crash looks systemic.
run-control
by tycenjmccann
Cancel, restart, or start workflow runs on an explicit user request ("kill that run", "cancel it and start over with X", "kick off a workflow to..."). Covers resolving WHICH run the user means, the cancel/start toolkit calls, and carrying input over to a fresh run. CHAT mode only — never autonomous.
setup
by tycenjmccann
Guided AgentCore Hub setup. Asks the user which modules they want, generates .env.local, and runs only the deploy scripts those modules need — with a verification gate between phases.