Content
67%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 dense, operational, and highly actionable: a decision procedure, concrete command templates, explicit error-recovery behavior, and a clear submit-then-stop protocol, with no filler or background explanation. Its main weaknesses are redundancy — chain/continuation routing rules are stated three or more times across the guardrails — and a long undifferentiated bullet list that would benefit from grouping. No bundle files are needed or present, and disclosure within the single file is adequate.
Suggestions
Consolidate the --chain/continuation-routing rules that currently appear in the Decision Card and again across three or more guardrail bullets into one authoritative 'Chain rules' group.
Group the 25-bullet Guardrails list into labeled clusters (chain routing, artifact flags, corrections/diagnostics, submit behavior) so the relevant rule can be located without scanning the whole list.
State the $TARGET/$MESSAGE substitution convention and the role of the literal `command` prefix once so the invocation templates are unambiguously copy-paste executable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is telegraphic and free of padding or explanations of known concepts, but the ~25-bullet Guardrails flat list repeats --chain and continuation-routing rules across several bullets (e.g., the chain guidance in the Decision Card reappears in the bullets at 'Do not probe --chain', 'Plain nested ask...', and 'If the current task is a CCB result-chain continuation...'). Mostly efficient, but noticeably could be tightened — anchor 3. | 3 / 5 |
Actionability | Two concrete heredoc invocation templates with explicit flag placement, plus exact command names for diagnostics (ask get, pend, watch, ping) and correction (ccb followup <active_job_id> --message "<correction>") make the guidance mostly executable. Minor gaps: the literal `command` prefix and $TARGET/$MESSAGE substitution conventions are left implicit, so it is not fully copy-paste ready — anchor 4. | 4 / 5 |
Workflow Clarity | The numbered Decision Card sequences the pre-ask decisions, and there is an explicit error-recovery checkpoint ("If CCB says 'ask --chain requires an active parent job', retry once with plain ask") plus a clear terminal rule ("After submit, stop. Do not wait... do not poll"). It falls short of anchor 5 because validation covers only one error case and the remaining recovery guidance is scattered across guardrail bullets rather than forming a coherent loop. | 4 / 5 |
Progressive Disclosure | There are no bundle files (references/, scripts/, assets/ absent), and the skill is a single self-contained file with well-labeled sections (Decision Card, Guardrails, invocation). Structure is good, but the body runs ~85 lines with an ungrouped 25-bullet guardrail list that could be subgrouped (chain rules, artifact rules, diagnostics) — a minor organization gap, anchor 4. | 4 / 5 |
Total | 15 / 20 Passed |