Content
75%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an efficient, well-sequenced instruction skill: copy-paste clarification and adversarial-review prompts, a concrete PRD structure, and explicit phase gating with a user-input checkpoint. Weaknesses are minor — a validation step (self-score) with no threshold or revise-if-low loop, a slightly vague research phase, and a dangling reference to a `prd.md` command not shipped in the bundle.
Suggestions
Add a feedback loop to Phase 3: define a self-score threshold and instruct revision (e.g., "if any category scores below 70%, revise that section before saving").
Resolve the dangling `prd.md` reference — either inline the dispatch syntax for the adversarial review or remove the pointer since the file is not in the skill bundle.
Tighten Phase 1 by specifying what to extract from each search (e.g., "market context: existing solutions and known gaps") and give Phase 4 a default filename convention.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is compact and phase-structured with little padding; the Phase 2 PRD outline and 100-point scoring breakdown are tight and earn their tokens. Minor trimmable bits remain — the "STOP - SKILL ALREADY LOADED" banner and repeated emphasis lines like "**WAIT for user response...**" and "Do NOT over-research" — keeping it at anchor 4 rather than the fully lean anchor 5. | 4 / 5 |
Actionability | Mostly executable guidance: the Phase 0 clarification question block is copy-paste ready, the adversarial-review prompt is given verbatim in a quotable block, and the PRD section structure is concrete (FR-001 format, P0/P1/P2, 2-3 personas). Minor gaps: Phase 1's "one for domain/market context" is vague, the 100-point framework has no stated pass threshold or action if the score is low, and Phase 4 does not specify a default filename pattern. | 4 / 5 |
Workflow Clarity | A clear six-phase sequence (clarify → research → write → adversarial review → self-score → save) with an explicit user-input checkpoint ("WAIT for user response") and skip conditions (--fast). It stops short of anchor 5 because the self-score step has no feedback loop — no threshold, revision step, or checklist — and Phase 2.5 relies on an external `prd.md` command whose dispatch syntax is not in this file. | 4 / 5 |
Progressive Disclosure | No bundle files exist, and the single-file body is well organized with clearly labeled phases, so structure is appropriate. The reference to "See `prd.md` command for full dispatch syntax" points to a file that is not present in the skill bundle, an unresolved navigation pointer that keeps it below the fully clean organization of anchor 5. | 4 / 5 |
Total | 16 / 20 Passed |