Conciseness

Every section carries non-obvious, project-specific knowledge — past wrong findings, the median/mean skew numbers (3.6M vs 15.4M), the `explore_share` null caveat — with no padding and no explanation of concepts Claude already knows. The two anecdote paragraphs earn their tokens by encoding exactly how prior analyses failed.

Actionability

Fully executable, copy-paste-ready commands throughout: `bun run --cwd tools/pr-metrics report -- --coverage`, `--waste`, `--json`, and the `gh pr diff … --patch | grep -E …` dependency-bump check, plus concrete output requirements for the written answer.

Workflow Clarity

Clear sequence (Step 0 coverage → Step 1 run what you need → Step 2 report) with explicit checkpoints: read and write down three coverage facts before continuing, exclude `null`-edit cards and report the exclusion count, state unrated data at the top, and seek a discriminator before accepting a correlation. This is an analysis skill, so the destructive/batch cap does not apply.

Progressive Disclosure

Well-organized self-contained sections with clearly signaled external pointers (`tools/pr-metrics/README.md`, `wiki/observability/session-metrics.md`, `queries.sql`, the `retro`/`pick-agent` skills) in a dedicated "What consumes this" section. It scores below 5 because the ~120-line body inlines all operational detail with no bundle split, and sibling-skill pointers are names rather than paths.

Specificity

Names several concrete actions — "running the report over `.claude/metrics/`", "interpreting it without the traps", "reporting what is actually supported" — but does not enumerate the individual analyses or traps, leaving minor gaps in coverage.

Completeness

Explicitly answers both: what ("running the report over `.claude/metrics/`, interpreting it without the traps…, and reporting what is actually supported") and when ("Use when reading the session-performance cards… Also the step before proposing any change to how work is dispatched") with concrete trigger phrases.

Trigger Term Quality

Good natural keyword coverage ("session-performance cards", "agent effort", ".claude/metrics", "how work is dispatched") that a user of this repo would plausibly say, though common variations like spend/cost/token usage are missing.

Distinctiveness Conflict Risk

The session-card corpus-analysis niche is distinct, but the closely related per-session `retro` skill is a minor overlap risk that the description itself does not disambiguate (only the body does).