Conciseness

The body is lean and assumes Claude's competence — "Verify the real artifact. Do not infer from proxies, self-reports, or 'it compiles.'" — with no explanation of known concepts; the only example (the manifest JSON) teaches a project-specific format Claude could not already know. Every token earns its place.

Actionability

Guidance is fully concrete: an exact output path ('.factory/evidence/<ticket>/<task>/manifest.json'), a copy-paste-ready manifest example with command, cwd, exitCode, and artifact, a hard acceptance rule ("A criterion without a command and a zero exit code is not proof"), and a specific UI variant ("the Surf screenshot and the read-back of the rendered page, taken against the worktree's own server").

Workflow Clarity

This is a simple, single-purpose skill and the single action is unambiguous: check the real artifact, then record evidence. It includes an explicit validation standard (zero exit code) and an error-recovery heuristic ("When a check fails, suspect the observation method before suspecting the product"), so the simple-skill exception applies cleanly.

Progressive Disclosure

The skill is 37 lines, needs no external references (none exist and none are referenced), and is well organized as principle → verification checklist → evidence format → commit rule. Per the under-50-line guideline, this earns the top score without section headers or bundle files.

Specificity

The description states one crisp concrete action — "Prove a task against the real artifact before calling it done" — and scopes it with trigger conditions rather than listing distinct capabilities. It matches the anchor 'names domain and 1-2 concrete actions, but not comprehensive'; it is not 4 because it does not enumerate several specific actions.

Completeness

It explicitly answers what ("Prove a task against the real artifact before calling it done") and when ("Use during factory-work, before ready_for_review, and whenever a check only shows that code compiles or a file changed") with concrete trigger phrases, matching the top anchor exactly.

Trigger Term Quality

Triggers like "whenever a check only shows that code compiles or a file changed" and "before ready_for_review" are natural phrases in this workflow, giving good keyword coverage. It falls short of 5 because common synonyms such as 'verify', 'test', or 'evidence' are absent and coverage leans on project-specific jargon ('factory-work').

Distinctiveness Conflict Risk

The evidence-manifest niche with triggers tied to 'factory-work' and 'ready_for_review' is mostly distinct, but 'prove it works' overlaps slightly with generic run/verify-type skills, so minor overlap risk keeps it at 4 rather than 5.