Content
85%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, well-structured investigative skill: an explicit three-phase workflow (route → investigate under rules → validate against a closing checklist) with strong progressive disclosure into four genuine reference files. The two dimensions at 4 reflect minor trimmable redundancy between the rules and the closing checklist, and the absence of any worked example in the body itself.
Suggestions
Trim the overlap between 'Rules for every investigation' and the closing checklist: checklist items 6–7 restate the reference table rows and rule 2, so cutting the checklist to non-overlapping verification items would make every token earn its place.
Add one compact worked example in the body — e.g. a filled-in provenance note line such as `alias prod points to version 12 <- lambda get-alias output, 2026-11-08` — to make the evidence-tracking rule copy-paste concrete without expanding the reference files.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and prescriptive with zero explanation of concepts Claude already knows — every rule adds domain-specific discipline (e.g. "Keep a short provenance note per claim (`claim <- artifact, record or query`)"), and details like "airport-code regions such as PDX = us-west-2, IAD = us-east-1" are genuinely non-obvious. It is not a 5 because closing-checklist items 6–7 restate the routing-table rows and rule 2 ("claims about live configuration come from the deployed artifact, not the latest stored version"), a minor duplication that could be trimmed. | 4 / 5 |
Actionability | As an instruction-only skill it gives concrete, copy-ready guidance: the provenance notation "`claim <- artifact, record or query`", the tactic "list resources, log groups and data stores in the likely regions and search for the identifiers in the request", and the directive "report unique entities separately from raw records". It is not a 5 because the body contains no worked example or sample command — the command-level specifics are delegated to the reference files — leaving minor gaps against the 'specific examples cover the common cases' anchor. | 4 / 5 |
Workflow Clarity | The sequence is explicit and phased: pick the reference first ("Read the reference that matches the question before you start"), apply the five rules during the investigation, then run a dedicated 10-item closing checklist ("Before submitting your analysis") that includes a re-check feedback loop ("reread your conclusion against each piece of evidence you collected and removed any statement that contradicts it"). This matches the anchor-5 pattern of clear sequence, explicit validation steps, and a checklist for a complex process; the skill is read-only, so the destructive-operation cap does not apply. | 5 / 5 |
Progressive Disclosure | The body is a lean overview that routes to four real, verified reference files via a question-type table ("What a deployed system currently does... → references/deployed-artifacts.md"), with all referenced paths present on disk. The references are one level deep — sibling links only, no nested chains — and content is appropriately split between overview and detail, matching the anchor-5 structure of well-signaled navigation. | 5 / 5 |
Total | 18 / 20 Passed |