Content
93%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.
An exemplary lean skill body: concrete tool calls, a copy-paste-ready context-capture template, explicit resume mechanics, and a sharp "What NOT to do" section guarding against the main failure mode. The only meaningful gap is handling the outcome of the duplicate-check search in the create workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and operational throughout — tool signatures, a compact resumable-context template, and short prohibition notes. Every parenthetical (provenance flag, status-move tools, Deeplake filesystem) carries real guidance rather than padding; nothing explains concepts Claude already knows. | 5 / 5 |
Actionability | Fully executable guidance: exact tool call signatures (hivemind_goal_add({ text })), a copy-paste-ready multi-line example showing the full context package, and the explicit memory path pattern "memory/goal/<owner>/opened/<goal_id>.md" for resuming. | 5 / 5 |
Workflow Clarity | The three workflows (express a goal, capture a task, resume a task) are clearly numbered with checkpoints — duplicate-prevention search first, capturing the returned goal_id, and confirming to the user. Minor gap: the pre-creation search surfaces an existing related goal but gives no instruction on what to do when one is found, and resume has no fallback if the search misses. | 4 / 5 |
Progressive Disclosure | At ~48 lines with no bundle files (no references/, scripts/, or assets/ exist), this fits the rubric's simple-skill exception: well-organized sections with no need for external references score 5. | 5 / 5 |
Total | 19 / 20 Passed |