Conciseness

The body is two lines with zero padding or over-explanation: a provenance comment and 'Load and follow the shared skill at ~/.agents/skills/flywheel/SKILL.md'. Every token earns its place; nothing could be trimmed.

Actionability

The only guidance is a redirect to an external path; there are no steps, commands, or examples for any of the promised tasks (define, review, repair, set up a flywheel). A specific path is given so it is not entirely vague, but the executable substance is entirely delegated, fitting 'minimal concrete guidance; missing the specific steps to execute'.

Workflow Clarity

The single directive is unambiguous (load and follow the referenced skill), but there is no validation that the target file exists, no fallback if it is missing, and no sequence for the multi-step flywheel tasks the description promises. It lands at 'steps present but checkpoints missing or implicit' rather than higher.

Progressive Disclosure

The reference is clearly signaled and one level deep, but the body contains no overview content at all and points outside the skill bundle ('~/.agents/skills/flywheel/SKILL.md', plus a 'scripts/agent-harnesses.py' comment for a file not present in the bundle), so navigation depends on an unverified external dependency. This fits 'some structure; references present but organization is incomplete' rather than a well-split overview.

Specificity

The description lists multiple concrete actions covering the full lifecycle: 'Define, review, repair, or set up', 'records agent sessions', 'learns from them by analysis or by replaying them on better models', 'applies the lessons to instructions, skills, hooks, and memory', 'prunes scaffolding', and 'tunes the loop itself'. This matches the comprehensive-coverage anchor; no meaningful capability gaps remain.

Completeness

Both halves are explicit: the 'what' is the detailed first clause enumerating the loop's operations, and the 'when' opens with 'Use when a repo documents a flywheel (docs/flywheel.md) that needs review or repair, when a repo has no learning loop..., or when the user asks how agents should learn across sessions'. This matches the anchor requiring concrete trigger phrases for both.

Trigger Term Quality

Natural user phrasing is comprehensively covered with synonyms and concrete artifacts: 'flywheel (docs/flywheel.md)', 'learning loop', 'learn across sessions', 'keep memory', 'run retros or evals on recorded sessions', and 'remove harness scaffolding as models improve'. These are phrases a user would naturally say when needing this skill.

Distinctiveness Conflict Risk

The agent-harness learning-loop niche is clear and mostly distinct, but trigger phrases like 'keep memory' and 'run retros or evals' carry minor overlap risk with general memory-management or eval skills. It fits 'mostly distinct; minor overlap risk with closely related skills' rather than the minimal-conflict anchor.