Content
87%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 exceptionally lean, actionable instruction skill: concrete endpoints, an exhaustive decision taxonomy, useful stuck-state heuristics, and a clear done-bar. The one substantive weakness is the absence of validation/feedback steps for a batch pass that includes destructive actions (close/cancel), which caps workflow clarity.
Suggestions
Add an explicit pass sequence (1. fetch inbox-lite, 2. fetch heartbeat-context per candidate, 3. classify, 4. act, 5. report) so the workflow order is stated, not inferred.
Add a validation checkpoint after destructive or status-changing actions (e.g. re-fetch the issue or confirm the status/comment took effect before moving to the next issue) to satisfy the batch-operation feedback-loop requirement.
Include a safety check before Close verdicts, such as confirming no child issues remain open or the parent's acceptance criteria are met.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every section is operational — endpoints, statuses, classification rules, and a don't-do list — with no concept explanations Claude already knows and no padding. "Lean and efficient; every token earns its place" fits exactly; the 4 anchor would require over-explanation that is not present. | 5 / 5 |
Actionability | The body gives exact API endpoints ("GET /api/agents/me/inbox-lite", "GET /api/issues/{issueId}/heartbeat-context"), exact status transitions, exact artifact types ("request_confirmation", "ask_user_questions", "request_board_approval"), and a concrete per-issue output format — fully executable instruction-only guidance per the rubric's code_vs_instruction note. | 5 / 5 |
Workflow Clarity | Per-issue classification rules and heuristics are clear, but the pass-level sequence (fetch inbox → fetch heartbeat context → classify → act → report) is implied by section order rather than sequenced, and the batch operation includes destructive verdicts (close/cancel) with no validation or verification step after actions — the rubric caps batch/destructive workflows without validation at 3. | 3 / 5 |
Progressive Disclosure | No bundle files exist and none are needed; the ~59-line body contains only overview-appropriate content organized into clearly titled sections (when to use / not use, inputs, decision, heuristics, don't-dos, output) with easy navigation, satisfying the well-organized-sections standard. Not a 4 because there are no organization gaps or misplaced content. | 5 / 5 |
Total | 18 / 20 Passed |