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.
The body is a dense, telegraphic rulebook with genuinely executable command forms, a well-sequenced decision procedure, and explicit error-recovery guidance — well above typical skill bodies. Its main weaknesses are moderate redundancy across guardrail bullets and no worked example instantiating the placeholders, plus a Guardrails section long enough that it could live partly in a reference file.
Suggestions
Consolidate the three near-duplicate --chain dependency rules into a single rule to trim tokens without losing coverage.
Add one concrete worked example (real target and message text) so the bash form is copy-paste runnable as-is.
Move the lineage/continuation edge-case rules (the CCB routing bullets) into a references file linked from a short "Edge cases: see LINEAGE.md" pointer.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is telegraphic and assumes competence throughout (e.g., "Automatic spill for text over 4 KiB is a fallback, not the primary rule", "Artifact flags are orthogonal to --chain, --silence, and --compact"), with no explanation of concepts Claude already knows. It falls short of 5 because the Guardrails list contains near-duplicative rules — the --chain dependency guidance appears at "Do not probe `--chain`...", "Never add `--chain` merely to make a rejected plain ask succeed", and "Plain nested `ask` from an active CCB task is rejected" — that could be consolidated. | 4 / 5 |
Actionability | Concrete, executable guidance dominates: two copy-paste bash forms ("command ask \"$TARGET\" <<'EOF' ... "), the exact recovery command "ccb followup <active_job_id> --message \"<correction>\"", and a step-by-step Decision Card mapping each intent to a specific flag. It is not 5 because only two of the many flag combinations are shown as ready-to-run examples, and placeholders like $TARGET/$MESSAGE are not instantiated with a worked example. | 4 / 5 |
Workflow Clarity | The Decision Card gives a clearly sequenced pre-submit procedure (need delegation? -> dependency gate -> result intent -> request fidelity), and guardrails supply explicit error recovery ("If CCB says `ask --chain requires an active parent job`, retry once with plain `ask`") plus a definitive post-submit step ("After submit, stop"). It is not 5 because there is no checkpoint confirming the submission succeeded, and several guardrails (lineage routing, continuation rules) are stated as assertions rather than ordered steps for the situation they cover. | 4 / 5 |
Progressive Disclosure | The body has clean section structure (Decision Card, Guardrails, command examples) with no dangling references and no nested-reference indirection; every rule is directly actionable from the single file, and no bundle files exist to misroute. It is not 5 because the ~60-line Guardrails rule list is dense enough that a reference file (e.g., edge-case/lineage rules) would keep SKILL.md closer to an overview, leaving minor organization gaps. | 4 / 5 |
Total | 16 / 20 Passed |