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 lean, imperative, project-grounded workflow: every line assumes Claude's competence and the commands are nearly copy-paste ready, with explicit anti-fabrication and cost-safety guardrails. The remaining gaps are minor — incomplete subcommand invocations, no explicit failure-recovery loop, and all referenced material living outside the skill bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every sentence carries project-specific, non-obvious constraints ('do not silently switch provider or relax permissions', 'Do not fabricate AR/IC/drawdown', 'Do not launch recursive Skills from a model-provider subprocess') with zero padding and no explanation of concepts Claude already knows. The ~16-line body is lean and every token earns its place, matching anchor 5 exactly. | 5 / 5 |
Actionability | The body gives copy-paste-ready commands ('python egs/skills.py preview --config <task.yaml>', 'python egs/skills.py run --config <task.yaml> --format jsonl') and a concrete template pointer ('Read `egs/skills_examples/quant-backtest-review.yaml`'), so most guidance is executable. Minor gaps remain: 'providers --probe' and the 'show', 'events', 'files', 'cancel' operations are not spelled out as full invocations, which keeps it below the fully-executable anchor 5. | 4 / 5 |
Workflow Clarity | Five clearly sequenced numbered steps include an explicit pre-run checkpoint (preview, with '--probe' when 'backend availability is uncertain') and an error path for missing evidence ('report the missing fields and retain failed-run evidence instead of repeating paid calls'). However, there is no explicit fix-and-retry feedback loop if preview or the run fails (only 'cancel' is mentioned), so it fits 'clear sequence with most checkpoints present; minor validation gaps' rather than anchor 5. | 4 / 5 |
Progressive Disclosure | The short body is a clear overview that defers detail to a well-signaled, one-level-deep reference ('See [the shared execution guide](../../../docs/usage/agent_cli_skills.md) for provider limits, data fields and research gates') plus the example YAML, so navigation is easy. It falls short of anchor 5 because no bundle files ship with the skill (no references/, scripts/, or assets/) and the referenced guide lives outside the skill directory, making navigation entirely dependent on the host project. | 4 / 5 |
Total | 17 / 20 Passed |