Conciseness

Dense and skill-specific throughout (16-lens table, S0–S8 pipeline in four lines, honest limits, anti-patterns) with no explanation of concepts Claude already knows. Not 5: the closing bash block ("python3 scripts/route.py --goal discovery --respondent customer --n 3") repeats commands already given in the Workflow steps, and the pilot-run guidance is restated in both "Exercise status" and "Honest limits" — minor trims available. Not 3: there is no padded or unnecessary explanation.

Actionability

Every workflow step carries an exact command with flags ("number_lines.py transcript.(txt|docx|srt|vtt)", "verify_quotes.py --transcript t_nl.txt --claims claims.json", "score_insights.py nuggets.json --k 3") and the bash block is copy-paste runnable via --sample. Not 4: the concrete invocations with realistic arguments cover the common cases end to end.

Workflow Clarity

S0–S8 are sequenced in order with explicit validation gates: quote verification and entailment in S2, "flags cells where independent runs disagree" in S3, "n < k interviews is a pilot ... synthesis yields watchlist, not insight", and flagged cells routed to a human ("Auto-resolving a disagreement ... the model never picks the winner"). Not 4: validation checkpoints and error-recovery routing are explicit, not implicit.

Progressive Disclosure

Each stage's detail sits one level deep behind a purpose-stated pointer ("Full routing matrix ... references/intake.md", "Index with stage mapping: references/pipeline.md", "references/ethics.md — consent, de-identification, and the lenses that need extra care"), keeping SKILL.md a genuine overview. Not 4: the split is clean and every reference is signaled with what it contains. Caveat: no references/, scripts/, or assets/ files were present in the review bundle, so the referenced paths could not be verified against actual files.

Specificity

"Maps a single interview transcript into a verbatim-grounded, lens-coded structure and checks whether analytic conclusions hold across independent re-runs", plus "Every conclusion must carry a verbatim quote verified against the source text; unstable conclusions are flagged for a human" — multiple concrete actions covering the skill's full capability set (map/code, quote verification, re-run stability check, flagging, cross-interview synthesis). Not 4: the action coverage is comprehensive rather than having 'minor gaps'.

Completeness

Explicit what ("Maps a single interview transcript into a verbatim-grounded, lens-coded structure...") and explicit when ("Use when the user wants to code, map, interpret, or make sense of an interview transcript...") with concrete trigger phrases, plus negative scope ("Not for generating new interviews, not for plain audio transcription..."). Not 4: the when-clause is fully explicit and concrete, not merely present.

Trigger Term Quality

Comprehensive natural phrasing: "code, map, interpret, or make sense of an interview transcript" plus the full synonym family "depth interview, custdev/JTBD/expert/usability/exit/win-loss/focus-group session", and even "even if they don't name a method". Not 4: no natural variant a user would say is missing; the description enumerates the whole range of interview-type names.

Distinctiveness Conflict Risk

"Not for ... research-method or sample-size planning above the transcript (see product-research)" draws an explicit boundary with the nearest neighboring skill, and the transcript-level coding niche is distinct. Not 4: no minor overlap risk remains — the boundary is stated outright rather than implied.