runtime-log-analysis

# Skill: Runtime Log Analysis > Sentinel: `TASKBOARD-SKILL-RUNTIME-LOG-ANALYSIS`. Echo this string in your first reply when the orchestrator probes whether the skill loaded. Turn the **product runtime observability** stream into a reviewable narrative about how the *built software* actually behaved during a run. This skill is the analysis side of the contract documented in [`docs/operations/runtime/observability.md`](../../../docs/operations/runtime/observability.md) and the capture layout in [`docs/operations/runtime/log-capture.md`](../../../docs/operations/runtime/log-capture.md). ## When to invoke User-triggered only. Typical entry points: - A user clicks a project-level "Runtime log analysis" button (or runs the skill directly from a CLI session). - A reviewer wants to know what the app did during the just-finished run, not what the agent said about it. - A regression hunt: pair a green Playwright run with the runtime stream from the same job and check whether tests passed while the app silently logged errors. Do **not** run this skill automatically from a hook, watcher, or auto-mode loop. The runtime stream is read-only output; analysis is a deliberate user action. ## Hard constraints These are policy, not preference. Violating any is a defect. 1. **Read-only.** Do not modify any source file, prompt, schema, or job folder. Do not move jobs between lanes. Do not call task-creation endpoints. 2. **No silent follow-ups.** A finding may be paired with...