Content
75%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 well-engineered skill body: table-driven, fact-dense, with a clear six-step workflow, explicit validation and cutover gating, and an exemplary reference routing structure that is one level deep and matches the actual bundle. The remaining gaps are moderate — repeated key facts that cost tokens, deferred executable command sequences, no explicit rollback path during cutover, and an orphaned assets/sqs-processor/template.yaml that no file links to.
Suggestions
Consolidate the repeated 90-minute/5400s timeout and 3-minimum-execution-environments facts into their primary sections plus the limits table, and replace the other recurrences with a short pointer to those sections to reclaim tokens.
Link the assets/sqs-processor/template.yaml example from SKILL.md (e.g., in the Files table or the infrastructure row of the routing table) so the bundle's ready-made template is discoverable.
Add an explicit rollback/abort checkpoint to Step 6 (e.g., 'if throttle rate exceeds 1% or costs regress during the shift, revert traffic to the Lambda Event Function and consult troubleshooting.md').
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and factual — decision tables, numeric limits, pricing components — with no padding explaining concepts Claude already knows; the SOCI and 90-minute-timeout detail is genuinely non-obvious knowledge that earns its tokens. Not 5: several key facts are repeated across sections (the 5400s/90-min timeout appears in the decision table, Step 1, Step 3, the long-duration section, and the limits table; the 3-environment minimum and 'set MaxVCpuCount explicitly' each recur ~4 times), which could be consolidated; not 3: the repetition is deliberate emphasis, not unnecessary explanation. | 4 / 5 |
Actionability | Guidance is concrete and executable: a specific CLI command ("update-function-configuration --timeout 5400"), SAM/CFN property ("Timeout: 5400"), exact memory-to-vCPU ratios, per-runtime concurrency caps, pricing math ("EC2 instance hours + 15% management fee + $0.20/1M requests"), alarm thresholds ("throttle rate > 1% and CPU > 80%"), and a staged cutover ("10% → 50% → 100%"). Not 5: complete command sequences and templates are deferred to reference files rather than given inline, so the body alone is not copy-paste complete for the common setup cases; not 3: what is present is specific and executable, not pseudocode. | 4 / 5 |
Workflow Clarity | The six-step workflow (Assess → Cost comparison → Configure → Migrate → Infrastructure → Validate and cut over) is clearly sequenced, with a Step 1 signal-gathering checklist and explicit validation in Step 6 ("Deploy to a non-production environment first", monitor CloudWatch, gradual 10→50→100% shift, "Decommission ... once stable") plus inline error-recovery facts (stuck CREATING → missing NAT route). Not 5: there is no explicit rollback/abort path if throttling or cost regression appears mid-cutover — the recovery loop lives in troubleshooting.md rather than the workflow; not 3: validation checkpoints are explicit and present at the right points. | 4 / 5 |
Progressive Disclosure | Structure is strong: a routing table instructing "Read ONLY the single reference file that matches the user's task", six real one-level-deep reference files (verified on disk, none nested), a Files index with content summaries, and key facts kept inline for always-mention use. Not 5: the bundle contains assets/sqs-processor/template.yaml which is never mentioned in SKILL.md or any reference — an orphaned asset that is invisible to navigation; not 3: content split, signaling, and depth are otherwise exactly right. | 4 / 5 |
Total | 16 / 20 Passed |