Content
85%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 SKILL.md body is a highly actionable, well-sequenced orchestrator with strong validation gates and clean one-level-deep references to verified playbooks. Its main weakness is conciseness: command blocks and rules are restated across stages, adding tokens without new information.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is operationally dense (routing tables, gate commands, hard rules) but repeats material: the pandoc + word_guard zh command block recurs in Stages 8, 10, and 12, and the Chinese completion rule is restated twice; this puts it at 'mostly efficient with some unnecessary content' rather than the lean 4 anchor. | 3 / 5 |
Actionability | Guidance is copy-paste executable throughout: exact commands like "python scripts/progress_check.py paper_rewriting_output --gate intake", full pandoc invocation with bibliography flags, and specific output file names cover the common cases; not 4 because gaps are negligible. | 5 / 5 |
Workflow Clarity | A 12-stage sequence is explicitly numbered with a per-stage gate, a resume-first rule, an anti-skip rule, a loop rule, and a final hard gate with feedback loops ("If FAILED: return to intake"), giving clear validation checkpoints and error-recovery loops for the destructive/batch audit operations. | 5 / 5 |
Progressive Disclosure | The orchestrator is an overview that routes to one-level-deep references via two explicit tables (trigger→references/*.md, stage→references/*.md); all referenced playbooks (resume.md, audit.md, latex.md, contribution.md, etc.) were verified to exist, and detailed per-stage content is correctly offloaded rather than inlined. | 5 / 5 |
Total | 18 / 20 Passed |