Content
82%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 strong operational skill body: lean imperative rules, copy-paste-ready commands with verified bundle files, and a fail-closed procedure with explicit stop checkpoints. The main gaps are a somewhat redundant Invocation section, no post-spawn verification of worker outcomes, and a couple of references (label-contract.json, inlined filter rules) that are not cleanly disclosed.
Suggestions
Consolidate the write-prohibition rules repeated across the Activation card, Invocation, and Hard nos into one canonical location to tighten conciseness.
Add an explicit post-spawn verification step (e.g., confirm each slot's worker completed before reporting "spawned") to close the workflow validation gap for this batch operation.
Give assets/label-contract.json an explicit read trigger and link, and move the inlined fetch_queue.py/triage_state.py filter rules into a short reference file or the scripts' docstrings to sharpen progressive disclosure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, imperative, and explains no concepts Claude already knows, but boundary rules are restated across sections: "This scheduler and its slots never assign and never request reviewers, in any ORIGIN" repeats the Activation card's "This scheduler never comments, labels, assigns, or requests reviewers" and the Hard nos, and the Ownership-writes list mostly describes other pipelines ("PR review: request the reviewing user as reviewer"). This fits anchor 4 (efficient with minor trimmable instances); it is above 3 because the redundancy is limited to safety-boundary restatements rather than general padding, and below 5 because the Invocation section could be consolidated. | 4 / 5 |
Actionability | Guidance is fully executable: copy-paste commands "python <this-skill>/scripts/fetch_queue.py --kind pr --mode sweep --repo microsoft/apm > batch.json" and "python <this-skill>/scripts/triage_state.py < batch.json", the dispatch variant "--mode dispatch --number N", fallback flags "--records-json" and "--labels-json", and fully specified card, receipt, and queue-table templates with every column defined. All referenced bundle files (fetch_queue.py, triage_state.py, label-contract.json, fan-out-pool.md) exist, matching anchor 5's complete, common-case-covering commands; anchor 4's "minor gaps" do not apply. | 5 / 5 |
Workflow Clarity | The Procedure is a clear numbered sequence with explicit fail-closed checkpoints at each stage ("Missing -> stop" for probes, "A helper failure stops the run", "Missing table, missing column, or blank rationale -> stop. Do not spawn."), so the batch-operation validation requirement is satisfied and the anchor-3 cap does not apply. It is not 5 because there is no post-spawn verification of worker outcomes beyond "Print a final report from the table" and no error-recovery loop (failures stop the run rather than feeding back), leaving minor validation gaps per anchor 4. | 4 / 5 |
Progressive Disclosure | Structure is good: on-demand loading is clearly signaled ("Load [assets/fan-out-pool.md](assets/fan-out-pool.md)"), scripts are reached via concrete commands, all four bundle files are referenced and exist, and references are one level deep. It is not 5 because "assets/label-contract.json" is named as a canonical owner with no read trigger or link, and the Selection section inlines ~15 lines of the helpers' filter rules ("closed, merged, locked, empty, template-only; sweep-only draft and spam ... at most two per author, oldest first, cap 10") that arguably belong in a reference or the scripts themselves — minor organization gaps per anchor 4. | 4 / 5 |
Total | 17 / 20 Passed |