Content
77%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.
This is a highly actionable, operationally rigorous skill: concrete commands, real cost figures, explicit validation gates, and genuine error-recovery loops for an expensive batch process. Its main costs are repetition of the same calibration and agent-choice rules across four-plus sections, and a progressive-disclosure posture that inlines guardrail/failure detail while pointing at resource files that are not present in the bundle.
Suggestions
De-duplicate the calibration recommendation and the ask-agent-choice rule: state each once (e.g. in the canonical workflow and guardrails respectively) and cross-reference instead of restating them in Goal, Expected outputs, Entry, Transitions, and multiple guardrails.
Move the full failure-and-recovery table and the detailed guardrail prose into the referenced resources/ files, keeping a short summary plus pointers in SKILL.md to cut its token footprint.
Ensure the referenced bundle files (resources/setup.md, scanning.md, pr-review.md, matchers.md, triage.md, config.md, and ../_shared/core/*.md) actually ship with the skill — the body relies on them but none are present in this bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with genuinely non-obvious operational detail (cost bands, failure-recovery table, two-job CI pattern), but the calibration recommendation "--limit 50 --concurrency 5" is repeated in at least four sections, the ask-agent-choice rule appears twice (Entry step 5 and Guardrail 13), and "Intent signature" largely duplicates "When to use", fitting the mostly-efficient-but-could-be-tightened anchor. | 3 / 5 |
Actionability | Every workflow ships copy-paste-ready commands (bunx deepsec process --limit 50 --concurrency 5, export --format md-dir, the process --diff PR invocation), concrete dollar cost bands, matcher hit-rate targets, and a specific failure-to-remediation table — fully executable guidance covering the common cases. | 5 / 5 |
Workflow Clarity | The sequence (Entry → Bootstrap → Calibrate → Full pass → PR mode → matchers → Resume) is explicit and includes real validation checkpoints for a costly batch operation: mandatory calibration before unbounded runs, cost extrapolation with explicit user go-ahead, quota-stop resume semantics, refusal handling, and revalidate-based verdicts — matching the clear-sequence-with-explicit-validation-and-recovery anchor. | 5 / 5 |
Progressive Disclosure | References are well-signaled and one level deep (a scoped References section, "loaded only when the scenario requires them"), but the referenced resources/*.md and ../_shared/core/*.md files are absent from the bundle, and substantial detail (13 guardrails, the full failure table, the resource-scope table) is inlined that belongs in those files — between the some-structure and good-structure anchors, held at 3 by the unverifiable references. | 3 / 5 |
Total | 16 / 20 Passed |