personal_agent

Use in the master session to integrate a ready PR — analyze (code-review + security-review), doc-drift check, merge, ask before deploy, verify live, close Linear, update MASTER_PLAN.

Use after /clear in the explore session (cc-explore) to rebuild full project situational awareness from durable sources — observer/strategist role, read-only, never actuate. All of master's vision, none of its hands; the deliberation seat where deep strategy/methodology thinking happens off master's context.

Use in the adr session (Opus) to explore an idea with the owner and — if it earns it — produce a complete ADR. Discuss/explore first, write, iterate with codex review, open ADR PR, then file sequenced implementation tickets. An ADR is a possible outcome, not a mandate. Never touches src/ or merges.

Use in the build session to ship a Linear FRE ticket from Approved to a master-ready PR (CI green + any bounce resolved) — fresh-start reset, plan with risk-tiered codex review (reviewer, not co-author), TDD, docs, PR, then self-complete via watcher/master pokes. Never merges or deploys.

Owner-invoked before resetting (/clear-ing) a session. Verifies a safe reset boundary, writes the session-delta artifact (LAST_SESSION.md) so the next session inherits this session's decisions, distills standing facts into memory, checkpoints MASTER_PLAN, and emits a go/no-go verdict. The safe wind-down bookend to prime-master.

Use after /clear in the master session to rebuild the guardian snapshot from durable sources — current state (1-7) then target (8) then process (9). Never from prior conversation.

Use when a task includes creating or changing a data visualization — a Kibana dashboard widget / Lens chart. Build the viz in the Kibana UI driven by Playwright, export a stable artifact, then run it through the create → scrutinize → iterate loop. NEVER hand-author Lens ndjson. Triggers — "create a widget", "add a dashboard panel", "Kibana chart", "Lens visualization", "visualize <telemetry>".

Pre-flight check and launch the recovery eval harness. Probes the primary chat endpoint AND the SLM/sub-agent endpoint before running so sub-agent 404s don't cascade into 600s primary timeouts. Use when the user asks to run, launch, or kick off an eval, the recovery harness, or `scripts/eval/recovery_harness.py`.

Create a Linear issue for the Personal Agent project with the correct team, state, and label policy. Use when the user asks to create, file, open, or draft a Linear issue/ticket. Do not invoke for reading or updating existing issues.