Content
70%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.
A well-structured orchestration skill with clear multi-loop workflows, explicit validation checkpoints, and concrete copy-paste artifacts. Its main weaknesses are conciseness (duplicated continuity blocks and repeated principles) and a broken templates/ reference on a mandatory onboarding step.
Suggestions
Remove the duplicated /loop prompt and cron JSON from the body, replacing them with a short pointer to references/agent-continuity.md, and consolidate the 'Research Discipline' principles with the sections that already state them (git protocol, inner-loop guidance) to cut redundancy.
Create the missing templates/ directory with research-state.yaml, research-log.md, and findings.md starter templates (or fix the references to point at real files), since line 58 makes it a mandatory initialization step that currently dead-ends.
Fix the dangling templates/progress-presentation.html link in references/progress-reporting.md and tighten vague fallbacks like 'reportlab, matplotlib, or similar' to one concrete default.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 411-line body is mostly efficient procedural content, but it duplicates the full /loop prompt and cron JSON that also appear in references/agent-continuity.md, and the 'Research Discipline' section re-states the git 'Lock before you run' rule and 'Never stop' principle already covered earlier, with some motivational padding ('Humans love seeing the upward curve'). | 3 / 5 |
Actionability | Highly actionable for an instruction skill: concrete workspace tree, git commit-message patterns, routing tables, a copy-paste /loop command and cron JSON, a trajectory JSON schema, and a sanity-check checklist; minor gaps include the broken templates/ reference and vague 'reportlab, matplotlib, or similar' fallbacks. | 4 / 5 |
Workflow Clarity | The Bootstrap → Inner Loop → Outer Loop → Finalize flow is clearly sequenced with explicit validation checkpoints (sanity-check list, lock-before-run git protocol, baseline reproduction) and feedback loops (stuck → literature/ideation, positive/negative branches, DEEPEN/BROADEN/PIVOT/CONCLUDE criteria). | 5 / 5 |
Progressive Disclosure | The three real references are clearly signaled and one-level-deep, but templates/ is referenced for the mandatory state-file initialization (line 58) and again in references/progress-reporting.md yet does not exist, and the agent-continuity setup is inlined in the body and duplicated in its reference rather than cleanly split. | 3 / 5 |
Total | 15 / 20 Passed |