Sentry skill ignores TELEMETRY.md before querying

[sentry-junior]

The sentry skill's workflow says to check TELEMETRY.md in the configured repo before forming queries, but in practice the agent skips this step and falls back to searching Sentry issues/errors instead of using the repo's telemetry recipes.

Observed behavior

• When asked about warden security-review findings (a usage/telemetry question), the agent searched Sentry error issues instead of loading TELEMETRY.md from the configured github.repo (getsentry/warden).
• The agent only consulted TELEMETRY.md after being explicitly prompted, at which point the correct query recipe was immediately obvious.
• The skill instruction "check the repo root for TELEMETRY.md before forming queries" was not followed on either the first or second turn.

Likely areas to investigate

• The sentry skill workflow step that says to check TELEMETRY.md may not be prominent enough relative to the default error-searching behavior. The agent defaulted to sentry issue list and sentry explore --dataset errors before considering logs, spans, or metrics recipes.
• The skill may need a stronger gate or ordering that forces TELEMETRY.md lookup before any Sentry CLI query when a github.repo is configured.
• Separately, the agent initially tried to resolve security-review as a Junior skill rather than recognizing it as a warden skill emitting telemetry — the configured repo context (getsentry/warden) should have been a stronger signal.

Evidence

• Thread: agent was asked "how many findings from the security-review skill in the last two weeks?"
• getsentry/warden/TELEMETRY.md contains a "Total findings for a skill" recipe and log completion patterns that answered the question directly once consulted.
• The correct answer (7 findings from Sentry logs) required the TELEMETRY.md recipes, not the errors dataset.

Action taken on behalf of David Cramer.