Content
63%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 an outstanding command reference — maximally concrete and executable with well-signaled links to external guides — but it is far over token budget for a SKILL.md, inlining hundreds of lines of edge-case behavioral detail with no bundle files to offload reference material into. Splitting per-agent and embedded-runtime detail into reference files would fix both the conciseness and disclosure weaknesses.
Suggestions
Move the agent registry table, embedded-hosts section, and session/queue ownership internals into reference files (e.g. references/agents.md, references/embedded.md) and keep one-line pointers in SKILL.md — this alone would roughly halve the body.
Trim changelog-style edge-case prose (guard-file semantics, model alias acknowledgement rules, -32603 replacement recovery) to the one sentence a user needs to act on, or delete it in favor of a link to the relevant docs page.
Add an explicit error-recovery loop to the main workflows (e.g. after sessions import, verify with 'sessions show' before prompting; on NO_SESSION, run 'sessions new' then retry) to raise workflow clarity.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~700 lines of dense edge-case behavioral specification — e.g. queue lease files and guard semantics ('/tmp/acpx-<home-hash>/<session-hash>.sock', 'Live guards never expire by age, and ambiguous guard state is preserved'), model aliasing acknowledgements, embedded session-record replacement rules ('may be replaced after an ACP internal load/resume error (-32603)'), and env allow/block layering. While none of it explains concepts Claude already knows (so not a 1), whole sections are release-note-level detail a SKILL.md consumer does not need, fitting anchor 2 ('noticeably verbose; several unnecessary explanations or padded sections') better than 3's 'some unnecessary explanation'. | 2 / 5 |
Actionability | Nearly everything is copy-paste executable: full command grammars, 'acpx codex "fix flaky tests"', 'acpx sessions prune --dry-run --older-than 7', a complete runnable TypeScript flow example with defineFlow/decision/decisionEdge, config key listings, and an NDJSON jq pipeline. It matches anchor 5 ('fully executable; copy-paste ready code or commands; specific examples cover the common cases'); it exceeds 4 because examples span the common cases across prompt, exec, sessions, and flows rather than having notable gaps. | 5 / 5 |
Workflow Clarity | Key multi-step processes are clearly sequenced: system-prompt override (create named session once, then select with -s), idempotent bootstrap (sessions ensure before every prompt), queue follow-ups with --no-wait, and prune with an explicit validation checkpoint (--dry-run preview before destructive deletion). It fits anchor 4 ('clear sequence with most checkpoints present; minor validation gaps') rather than 5, because most workflows are presented as reference recipes without explicit validate-then-proceed feedback loops (the dry-run prune is the only clear one), and flows/callback contracts lack error-recovery loops. | 4 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), so all content lives in one ~700-line SKILL.md. Section headers are clear and external guide links are well signaled (e.g. 'See the [OpenClaw guide](...)', 'See Watching sessions'), but large bodies of material that clearly belong in separate reference files — the 25-entry agent registry, embedded-host lifecycle rules, permission serialization semantics — are inlined. This matches anchor 3 ('some structure... content that should be separate is inline') better than 2, whose 'minimal structure' does not apply given the 20+ well-organized sections. | 3 / 5 |
Total | 14 / 20 Passed |