Content
81%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.
Highly actionable routing instructions with concrete per-host invocation syntax, explicit failure/limitation handling, and a firm output contract. The main weakness is redundancy: the 'never run the child CLI inline' rule is repeated four times and could be consolidated to one prominent statement plus one backstop reminder.
Suggestions
Consolidate the 'never run the child CLI inline' rule: state it once prominently in the opening and keep one backstop reminder in the Output contract, removing the restatements from the Cursor bullet and Step 2 preface.
Merge the 'When to use this skill vs ai-ide-runner' guidance into the existing scope-boundaries section (or reference the description's routing sentence) to avoid covering the same disambiguation twice.
In Step 2, trim the Claude Code entry to a single current tool name (the `Agent` tool) since listing both 'Agent / Task' invites ambiguity on which to invoke.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core rule ("You are NOT the courier — the `worker` subagent is") is restated four times: the opening paragraph, the Cursor bullet ("that would defeat the context-isolation point that motivates this skill"), the Codex bullet ("do not run the child CLI inline"), and the closing "Do not, under any circumstances, run `claude -p` / `codex exec` ...". The content is otherwise efficient, but this repetition goes beyond the 'minor instances of over-explanation' of the level-4 anchor, fitting 'mostly efficient but could be tightened'. | 3 / 5 |
Actionability | Fully executable per-host guidance: "use the `Agent` / `Task` tool with `subagent_type=worker`", the "`@worker <task prompt>`" mention syntax, "`spawn_agent` tool with `agent_type=\"worker\"` and `fork_turns=\"none\"`", a copy-paste worker prompt block with placeholders, and a concrete role-install check ("`ls \"${CODEX_HOME:-$HOME/.codex}/agents/\"`"). Common cases are covered with commands the agent can run directly. | 5 / 5 |
Workflow Clarity | A clear three-step sequence (parse intent → invoke worker → relay reply) with explicit precondition checks (Codex worker.toml presence via `ls`, the Cursor no-dispatch case) and error-recovery paths ("an empty listing means the role is not installed — say so, do not run the child CLI inline"; "pass that message through and stop"). The Output contract adds a MUST checklist for the final message. | 5 / 5 |
Progressive Disclosure | No bundle files exist, so all content lives in one file; it is well-sectioned (When to use vs ai-ide-runner, Workflow, Output contract, Scope boundaries) and nothing clearly warrants splitting. However, at ~105 lines it exceeds the under-50-line simple-skill exception that would allow a 5 on section organization alone, so it fits 'good structure; most content appropriately placed; minor organization gaps'. | 4 / 5 |
Total | 17 / 20 Passed |