Conciseness

The body is dense with non-obvious, environment-specific knowledge (exit 78 semantics, trunk-merge branch behavior, DEPOT_TOKEN overriding the saved login) and never explains concepts Claude already knows, but at ~380 lines it has some redundancy: the platform-outage guidance appears in both 'Rule out a platform outage first' and again in the Base rate section, and Depot retry is covered in both step 3 and the Backend-tests-on-Depot subsection. Anchor 4 ('minor instances of over-explanation that could be trimmed') fits better than anchor 5, since not every token earns its place.

Actionability

Guidance is fully executable: copy-paste commands with exact jq filters (the github-actions[bot] comment query, the check-runs API call for 'Backend CI on Depot / Test results'), concrete tool invocations (hogli ci:insights search "<error>", depot ci retry <run-id> --failed --workflow <workflow-id>), classification and reproduction tables mapping log signals to first actions, and a literal report template. This matches anchor 5 ('copy-paste ready... specific examples cover the common cases').

Workflow Clarity

The multi-step process is clearly sequenced (insights digest → find the failing run → read the CI report comment → classify → base rate → local repro → report) with explicit validation checkpoints: 'confirm from run data before reporting a failure as already fixed', 'Check the current queue run with gh before reporting it as active', and 'If it disagrees with the current logs, report the mismatch and use the logs for the cause'. Risky operations are gated behind explicit approval in Safety rules, so no destructive/batch cap applies. This matches anchor 5's explicit validation and error-recovery loops.

Progressive Disclosure

Sections are clearly headed and cross-skill handoffs are named, but the skill's own key reference — references/master-red-incident.md, the designated first step for the unattended workflow — does not exist in the bundle (there is no references/ directory at all), and ../../../docs/internal/ci-things-already-tried.md points outside the skill tree. Long operational detail (Backend tests on Depot CI, merge-queue branch semantics) is inlined in the single SKILL.md rather than split out. Anchor 3 ('references present but... content that should be separate is inline') fits; anchor 4 requires only 'minor organization gaps', which a broken primary reference is not.

Specificity

The description enumerates multiple concrete actions — "Debugs failing GitHub Actions CI runs for PostHog PRs, commits, and branches", "answers broad CI-health questions", "read-only inspection, failure classification, the smallest local reproduction with hogli, and safe reporting without rerunning CI or posting to GitHub" — with comprehensive coverage and no vague filler. Anchor 4 is too low because there are no gaps in the action coverage; this matches the comprehensive anchor 5.

Completeness

Both halves are explicit: the 'what' is stated up front (debugs failing runs, answers CI-health questions, then inspection, classification, reproduction, and safe reporting) and the 'when' is an explicit "Use when the user asks why CI is red, asks for the current CI or master status, or mentions a failing check..." clause with concrete trigger phrases, plus a separate unattended-mode trigger. This matches the anchor-5 example structure; anchor 4 ('when could be more explicit') is clearly weaker than what is written.

Trigger Term Quality

Trigger terms are comprehensive and natural, including quoted questions users would literally ask ("is CI red?", "is master green today?", "what's broken right now?") plus synonyms and domain-specific phrasings ("failing check", "GitHub Actions run", "Depot runner", "merge queue kick", "flaky test", "lint failure", "typecheck failure", "snapshot diff", "migration check", "generated types drift"). No common variation is missing, so anchor 4's 'a few natural terms missing' does not apply.

Distinctiveness Conflict Risk

The description is scoped to a clear niche ("PostHog PRs", "GitHub Actions", "Depot CI run", "hogli ci:insights", merge-queue kicks) with distinct trigger vocabulary, making it unlikely to fire for unrelated skills. Anchor 4's 'minor overlap risk with closely related skills' is not warranted since the PostHog/Depot/hogli vocabulary is highly specific.