Content
100%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.
An exemplary skill body: executable commands with concrete values, a decision-first mode-selection table, an operational guardrail checklist with validation and feedback steps, and clean one-level progressive disclosure to five real bundle files. No material weaknesses found.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and dense throughout: decision table, flag lists, and command blocks with no concept padding — every line carries version-specific operational detail Claude would not reliably know ("the `--dangerously-skip-permissions` dialog defaults to 'No, exit'", "Session resume is per-directory", "Slash commands don't work in `-p`"). Matches anchor 5; nothing to trim without losing information. | 5 / 5 |
Actionability | Fully executable, copy-paste-ready commands cover the common cases: setup (`npm install -g @anthropic-ai/claude-code && claude auth login`), the primary `terminal(...)` invocation with concrete values (`--allowedTools 'Read,Edit'`, `--max-turns 10`, `timeout=180`), and five variants including pipe-review, `--resume`, `--json-schema`, CI (`--bare`), and cost caps (`--max-budget-usd 0.50`). Matches anchor 5. | 5 / 5 |
Workflow Clarity | Explicit sequence — "Setup (once)" → "Platform check (before running anything)" → "Mode Selection (decide FIRST)" → mode sections → guardrail checklist — with validation steps ("verify with `claude auth status` / `claude doctor`", "Report back... use the JSON `result` + `git diff`"), a feedback loop ("monitor `/context`, use `/compact`"), and safety checks ("only run in trusted directories", "Don't kill slow sessions — capture progress first"). Matches anchor 5. | 5 / 5 |
Progressive Disclosure | A ~70-line overview with all detail pushed to five verified, one-level-deep reference files, each clearly signaled inline ("→ `references/flags.md`") and in a closing "read when" table. Structure mirrors the anchor-5 example; no nested reference chains and no content that should be externalized inlined in the body. | 5 / 5 |
Total | 20 / 20 Passed |