Content
92%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.
An excellent process skill body: a clearly sequenced bounded-turn workflow with explicit verification, error recovery, and state rules, fully concrete commands, and a properly split two-file bundle that is clearly signposted. The only room for improvement is trimming the small redundancy between the opening paragraph and the frontmatter description.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and skill-specific: tables for layers and header fields, terse rules ("Never claim a task another agent updated within the last 2 hours"), and no filler explaining concepts Claude already knows. It falls short of lean-perfect only because the opening paragraph ("The file is both the ticket and the handoff: whoever reads it can continue the work without the previous conversation") restates the handoff idea already given in the frontmatter description, and a few connective sentences could be trimmed. | 4 / 5 |
Actionability | Fully executable guidance throughout: concrete copy-ready commands (`ov grep '^(title|status|owner|updated):' -u <root>/<board> -x <root>/<board>/archive -o json`, `ov write <root>/<board>/<id>.md --from-file ./task.md --wait`, `date +%Y-%m-%dT%H:%M:%S%z`), a template file to copy, and specific error handling for every named failure (PERMISSION_DENIED fallback root, CONFLICT retry, ALREADY_EXISTS re-id, MCP-timeout read-back). Placeholders are all defined. | 5 / 5 |
Workflow Clarity | "One turn" gives a clear numbered sequence (Pick, Claim, Slice, Verify, Write back) with an explicit verification checkpoint ("Run the check that proves the slice... Only verified work goes into Progress"), feedback loops for error recovery (CONFLICT wait-and-retry, read-back after an MCP-timeout write), a checklist convention (first unchecked `Next` box is the next action), and invariant rules (never `done` with an unchecked `Goal` box). Since validation is present, the destructive/batch cap at 3 does not apply; this matches the top anchor. | 5 / 5 |
Progressive Disclosure | Scored against the actual bundle: both referenced files exist (references/task-template.md, 32 lines; references/loop.sh, 56 lines), are linked with clear markdown signals at the point of use ("Copy [references/task-template.md]...", "[references/loop.sh] runs one agent turn after another"), and are one level deep with no nested references. The executable detail (template, unattended loop) is appropriately split out while the core process stays in SKILL.md. | 5 / 5 |
Total | 19 / 20 Passed |