It avoids explaining concepts Claude already knows, but the ~350-line body inlines substantial mechanics (the Constants block, the full Resumable runs section, and the complete heartbeat stall-detection bash) that could be tightened or offloaded to references.

Provides copy-paste-ready invocations ("/idea-discovery \"$ARGUMENTS\""), concrete run_state.py / iteration_log.py commands with exact arguments, and explicit artifact paths — fully executable guidance throughout.

Stages 1–5 are clearly sequenced with two explicit gates (Gate 1 human checkpoint, Gate 2 writing checkpoint), a per-phase accept table, and a validation/feedback loop (re-validate done-but-unaccepted stages on resume, never accept on the executor's own say-so).

Shared-reference links (external-cadence.md, resumable-runs.md, output-versioning.md) are one-level-deep and clearly signaled, but no bundle files exist in references/scripts/assets and large blocks of content that belong in those references (resumable runs, heartbeat detection) are inlined rather than split out.

Lists multiple concrete pipeline stages — "idea discovery, experiments, and review all the way to a polished paper PDF" — rather than vague language, matching the multiple-specific-actions anchor.

Explicitly answers both what (full end-to-end research pipeline from direction to paper PDF) and when ("Use when user says...") with an explicit trigger clause.

Provides natural trigger phrases a user would actually say in both English and Chinese ("full pipeline", "end-to-end research", "全流程", "从找idea到投稿"), giving good coverage rather than jargon.

The niche is clearly the complete autonomous research lifecycle with distinct triggers like "full pipeline" and "全流程", making it unlikely to fire for a narrower single-stage skill.