What to build
Calibrate the silent-failure-hunter agent prompt (agents/silent-failure-hunter.md) to also surface lost-signal / observability gaps — fallback, early-return, or dev-only-log paths where a real event or error never reaches production observability (logs / Sentry / telemetry) — without inflating false positives.
This is the clearest behavioural-parity gap from the PR #5612 comparison: same agent name, same input, divergent output. The toolkit's silent-failure-hunter flagged that a cancelled-route transition which fails to complete loses the analytics page_view with no production trace; unic's silent-failure-hunter returned 0 findings. The agent is Markdown instructions to an LLM, so there is nothing to unit-test — verification is by example (re-run on the #5612 diff).
Acceptance criteria
Blocked by
None - can start immediately.
What to build
Calibrate the
silent-failure-hunteragent prompt (agents/silent-failure-hunter.md) to also surface lost-signal / observability gaps — fallback, early-return, or dev-only-log paths where a real event or error never reaches production observability (logs / Sentry / telemetry) — without inflating false positives.This is the clearest behavioural-parity gap from the PR #5612 comparison: same agent name, same input, divergent output. The toolkit's
silent-failure-hunterflagged that a cancelled-route transition which fails to complete loses the analyticspage_viewwith no production trace; unic'ssilent-failure-hunterreturned 0 findings. The agent is Markdown instructions to an LLM, so there is nothing to unit-test — verification is by example (re-run on the #5612 diff).Acceptance criteria
page_view/ observability finding; a before/after example is documented.Blocked by
None - can start immediately.