Conciseness

The ~30-line body is lean with zero padding: every sentence carries repo-specific contract knowledge Claude cannot infer (e.g., "The NaN allocation guard covers eligible FLA allocations in operator/module tests, including the single-rank CP graph and intra-card cache cases..."). No basic concepts are re-explained.

Actionability

Copy-paste-ready commands ("python scripts/find_dependent_tests.py fla/ops/kda/chunk.py", "FLA_CI_ENV=0 pytest tests/ops/test_kda.py -v") plus a concrete coverage-dimension table make the guidance mostly executable. Not 5 because how to concretely execute gradient checks, routing verification, and baseline reproduction is directed but not shown.

Workflow Clarity

A clear select-coverage -> run-affected-paths sequence exists with explicit validation checkpoints ("Inspect skips and warning-only comparisons before calling the run a pass", "reproduce failures on the unchanged baseline before labeling them pre-existing", "investigate a poisoned-run failure even when an ad hoc run passes"). Not 5 because the sequence is prose-embedded rather than an explicit ordered workflow with feedback loops.

Progressive Disclosure

The body is a concise overview with well-signaled, one-level-deep references: the four operator-math files (delta-rule, generalized-delta-rule, simple-gla, cp) all exist under references/ and are gated by "Read operator math only when needed". Good-overall-example structure (lean sections, deferred detail) is matched.

Specificity

"Select and run correctness coverage", "gradients, dispatch boundaries, and Triton addressing changes" list several concrete actions and areas; below 5 because other covered dimensions (state, layouts, routing fallback) are absent from the description, above 3 because more than 1-2 specific actions are named.

Completeness

The "what" is clear and concrete, but there is no "Use when..." clause or any equivalent explicit trigger guidance, which caps completeness at 3 per the judging guidelines. Below 4 because the "when" is entirely absent rather than merely less specific.

Trigger Term Quality

"correctness coverage", "FLA kernels", "gradients", "Triton addressing" are relevant keywords, but the natural terms users would most often say for this task ("tests", "testing", "verify", "pytest") are missing. Not 4 because the gap is more than a few synonyms; not 2 because the included terms are real, domain-appropriate trigger words.

Distinctiveness Conflict Risk

"FLA kernels and modules", "dispatch boundaries", "Triton addressing changes" carve a distinct niche with low conflict risk; only minor overlap with a sibling design-coverage skill keeps it below 5.