Content
88%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: concrete, executable guidance for both local and cloud channels with validation checkpoints and error-recovery loops at every destructive step. Its weaknesses are modest duplication (error codes explained twice, cloud-mode detail repeated between body and references) that costs tokens without adding information.
Suggestions
Explain BUILD_OUTPUT_NOT_FOUND / DEPENDENCY_NOT_INSTALLED once and reference the other occurrence (e.g., keep the error-recovery detail only in the Local workflow step), and cut the rhetorical Q&A sentence in 'Build & pipeline execution' to its conclusion about execution authority.
Move the cloud-mode two-phase zip and CloudApp upload step details into references/plan-and-apply.md's 'Cloud-hosted upload pipeline path' section, keeping only a brief channel-switch summary and routing link in SKILL.md.
Consolidate the 'Cloud mode' section's per-resource sub-sections into the existing reference flow to reduce the body's length and near-duplicate coverage of the build/plan/apply split.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational — tables, terse bullets, exact tool calls and error codes — with almost no explanation of concepts Claude already knows. Minor over-explanation keeps it from 5: BUILD_OUTPUT_NOT_FOUND and DEPENDENCY_NOT_INSTALLED are each explained twice (Local workflow and Build & pipeline execution), and the rhetorical "So the answer to 'can cloud mode run local CLI commands' is..." passage could be trimmed to its conclusion. | 4 / 5 |
Actionability | Guidance is fully executable: exact tool invocations with parameters (`deployBuild({ cwd, mode? })`, `deployApply` with `confirm=true`, `manageApps(action=deployApp, cosTimestamp=<unixTimestamp>, installCmd?, buildCmd?, deployCmd?)`), exact MCP action names, the strict schema warning for the COS triple, and named error codes with remediation. Not 4 because the concrete detail covers the common local and cloud cases completely. | 5 / 5 |
Workflow Clarity | The Local workflow is a clear 6-step build→plan→apply sequence with explicit validation checkpoints and feedback loops: read the plan's action classification before apply, stop on `conflict`, Deployment Gate declaration before any apply, and error-recovery loops (BUILD_OUTPUT_NOT_FOUND → run deployBuild then retry; DEPENDENCY_NOT_INSTALLED → run install first). The 'Minimum self-check' checklist reinforces it — this destructive/batch skill has validation throughout, satisfying the cap rule. | 5 / 5 |
Progressive Disclosure | Good structure: a Routing table maps user tasks to the three real one-level-deep reference files (all verified to exist, including the cited 'Cloud-hosted upload pipeline path' section in plan-and-apply.md), plus a Reference index. Not 5 because the SKILL.md itself inlines substantial detail that also lives in the references — the cloud-mode two-phase zip/CloudApp upload steps and build-executor tables could largely move to plan-and-apply.md, leaving the body a leaner overview. | 4 / 5 |
Total | 18 / 20 Passed |