Content
73%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.
A strong, highly actionable workflow document: every step carries executable commands, explicit prompts, and — rare for archive-style skills — mandatory post-sync verification with clean failure semantics for a destructive move. The main costs are redundancy between inline rules and the Guardrails section, and dense inline policy blocks that would sit better in a reference file.
Suggestions
Deduplicate the copy/verbatim rules: state them once either in step 1's instruction-contract paragraph or in Guardrails, not in both nearly verbatim.
Move the store-selection rules and the prompt-level context/guidance contracts into a one-level-deep reference file (e.g., references/store-and-instructions.md), keeping a one-line pointer in SKILL.md.
Replace "typically tasks.md" with the concrete path-resolution rule (e.g., read artifactPaths from the status JSON) so the task check is fully executable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The procedural guidance is dense and purposeful, but the instruction-contract paragraph ("These are prompt-level behavior contracts, not enforceable checks... do not copy their text verbatim") is restated nearly verbatim in the Guardrails section ("Never copy runtime context, operation guidance, or artifact-rule text verbatim"), and the store-selection paragraph is wordy. Fits anchor 3 — mostly efficient but could be tightened — rather than 4. | 3 / 5 |
Actionability | Concrete executable commands throughout ("openspec instructions archive --change \"<name>\" --json", "openspec status --change \"<name>\" --json", "mkdir -p", "mv") with exact prompt options and answer routing. Not score 5 because the tasks file is hedged ("typically tasks.md") and the spec merge is delegated to the openspec-sync-specs workflow rather than specified here. | 4 / 5 |
Workflow Clarity | Six clearly sequenced steps with explicit validation checkpoints and feedback loops for a destructive operation (mv of changeRoot): completion checks with user confirmation, mandatory post-sync re-verification ("each capability must now read as already synced"), and stop-on-failure handling ("report what differs and stop — do not archive"). Matches the anchor-5 example's validate → fix → retry pattern. | 5 / 5 |
Progressive Disclosure | Well-organized single-file skill (numbered steps, output template, guardrails) with no bundle files to reference. Not score 5: at ~180 lines the dense store-selection rules and prompt-level behavior contracts read as inline material that could live in a one-level-deep reference file. | 4 / 5 |
Total | 16 / 20 Passed |