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 a dense, well-sequenced orchestrator overview with genuine validation checkpoints and concrete CLI commands, and it mostly resists padding. Its main weaknesses are unresolvable, inconsistently-prefixed reference paths (with no reference files shipped in the bundle) and repeated authorization rules that inflate the token cost of an otherwise lean document.
Suggestions
Ship the referenced files in the bundle and use one consistent path convention — the same document is cited as both `comet-classic/reference/scripts.md` and `reference/scripts.md`, and `../comet/reference/host-check.md` points outside the skill.
Consolidate the authorization protocol ("ask only when a new decision is needed", existing-authority reuse) into the single decision-point section instead of restating it in sections 1, 2, and 3.
Split the overloaded multi-clause sentences (e.g., the state-transfer and restore-consent paragraph in section 1) into discrete steps so each instruction carries one condition.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is telegraphic and free of concept over-explanation — every line is instruction (e.g., "Only `auto_resume` enters automatically; for `ask_user` ask by `reason`"). It falls short of 5 because the authorization rule "ask only when a new decision is needed" is restated nearly verbatim across sections 1-3 and several overloaded run-on sentences (e.g., the state-transfer paragraph) could be tightened, fitting 'Efficient; minor instances... that could be trimmed'. | 4 / 5 |
Actionability | Concrete commands exist (`comet classic workspace resolve <change-name> --json`, `comet state select`, `comet resume-probe . --stdin --json`, `comet state restore <change-name> --confirmed`) plus an explicit phase-to-skill routing table, but many key procedures (host capability check, decision-point protocol, recovery procedure) are deferred to reference files that are not present in the bundle, and reference paths are inconsistently prefixed. This matches 'Some concrete guidance but incomplete; missing key details' rather than 4, where gaps would be minor. | 3 / 5 |
Workflow Clarity | A clear three-step sequence (resolve target/workspace, load current phase, complete authorized work) with a phase-to-skill table, and validation checkpoints are present: phase exit checks, Guards ("do not bypass phase Guards"), "Claim a test or build result only after actually running it", and archive-time verification of "actual Git state". Not 5 because checkpoints are scattered through dense prose rather than presented as explicit validate-then-proceed steps, leaving minor validation gaps. | 4 / 5 |
Progressive Disclosure | The overview-plus-triggered-references design is sound ("Read only sections needed for the current action", "Read other references only when their triggers above apply", one level deep), but the bundle contains no reference files at all, and the ~12 cited paths use three inconsistent conventions (`reference/...`, `comet-classic/reference/...`, `../comet/reference/...`), so navigation cannot be relied on. This fits 'Some structure but could be better organized' despite references being clearly signaled. | 3 / 5 |
Total | 14 / 20 Passed |