Conciseness

Mostly operational content rather than explanations of known concepts, but the per-phase "When AUTO_PROCEED=true (non-blocking)" / "When AUTO_PROCEED=false (blocking)" boilerplate in Phases 1, 2, and 4.5 duplicates the dedicated "Checkpoint execution rule" section, and the dense reviewer-slot preamble adds further repetition. Not 4 because the duplication is noticeable across multiple phases; not 2 because there is no padding about concepts Claude already knows.

Actionability

Gives concrete sub-skill invocations with exact flags ("/research-lit "$ARGUMENTS" — sources: all, gemini — composed: idea-stage/IDEA_REPORT.md"), exact state-record/gate command templates, and a full report template. Not 5 because commands rely on placeholders (<resolved-python>, <run_id>, <thread-or-trace-id>) the executor must resolve and some steps remain prose-level ("Brainstorm 8-12 concrete ideas via GPT-6-Astra xhigh"); not 3 because the large majority of guidance is directly executable.

Workflow Clarity

A clearly sequenced Phase 0 → 5.6 pipeline with explicit validation: the per-stage evidence gate with artifact locators, explicit error-recovery loops (negative verdict → leave phase done, gate BLOCKED, re-run the reviewer-bearing phase), and BLOCKED gate results folded into the final report rather than hidden. Feedback loops for the batch/destructive-adjacent operations (GPU pilots, state records) are all present.

Progressive Disclosure

Well-organized sections and clearly signaled links to ../shared-references/ protocols, but this is a single ~530-line file with no references/, scripts/, or assets/ bundle, and detailed contracts (evidence-gate CLI semantics, reviewer-slot routing, run-state commands) are inlined where they would fit better in reference files. Not 2 because navigation is easy and links are clearly labeled; not 4 because substantial content that belongs in separate files remains inline.

Specificity

"Full idea discovery pipeline to go from a broad research direction to validated, pilot-tested ideas" names a concrete start and end state but lists none of the pipeline's discrete actions (literature survey, brainstorm, novelty check, pilot experiments) that the body actually performs. Not 4 because no specific actions are enumerated; not 2 because it is far more concrete than a generic domain label like "Processes PDF files".

Completeness

Explicitly answers both what ("pipeline to go from a broad research direction to validated, pilot-tested ideas") and when ("Use when user says ...") with concrete trigger phrases, matching the anchor-5 pattern. Not 4 because the 'when' clause is fully explicit with multiple concrete triggers rather than a vague "Use when working with" formulation.

Trigger Term Quality

"Use when user says '找idea全流程', 'idea discovery pipeline', '从零开始找方向'" provides natural trigger phrases a user would plausibly say. Not 5 because common English variations ("find research ideas", "research topic selection", "come up with ideas") are missing; not 3 because the included phrases are natural and specific rather than generic jargon.

Distinctiveness Conflict Risk

"idea discovery pipeline" carves a clear niche with distinct trigger phrases unlikely to fire for unrelated skills. Not 5 because the skill lives in an ecosystem of closely related research workflow skills (/idea-creator, /research-pipeline referenced in the body), creating minor overlap risk within that family.