Content
100%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 excellent CLI-skill reference: dense, tool-specific tables and executable examples with zero filler; a clear three-step call flow; and a subprocess contract (ready marker, stdin EOF semantics, exit codes, JSON error envelopes) that doubles as validation and error-recovery guidance. Domain specifics are correctly deferred to eight real, one-level-deep reference files indexed with per-file coverage summaries.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every section carries lark-cli-specific knowledge Claude cannot already know (custom flags, the subprocess contract, pre-decoded payload behavior), delivered via dense tables and terse prose — e.g. the exit-code table and "any jq runtime error skips the event with a one-line WARN; the loop does not abort". No concept explanations or padding; matches the 5 anchor ('every token earns its place') rather than 4, which requires instances of over-explanation. | 5 / 5 |
Actionability | All examples are copy-paste executable ("lark-cli event consume im.message.receive_v1 --max-events 1 --timeout 30s --as bot", the multi-process background pattern with `wait`), the ready marker is given verbatim ("[event] ready event_key=<key>"), and failure handling names exact fields to branch on ("Parse `error.type` / `error.subtype` to branch... do not regex-match message text"). Fully executable per the 5 anchor with no pseudocode or missing steps. | 5 / 5 |
Workflow Clarity | A numbered call flow (list → schema → consume) plus a rigorous subprocess contract with explicit validation checkpoints: block on the stderr ready marker ("Do not fall back to `sleep`"), an exit-code/reason table, structured error envelopes whose `error.hint` gives "the recovery action", and completion semantics ("treat `reason: limit/timeout/signal` (all exit 0) as 'business completion'"). Self-correcting failures (unknown key/domain/param fail with the valid values listed inline) provide the error-recovery feedback loops of the 5 anchor; the 4 anchor's 'minor validation gaps' do not apply, and no destructive/batch cap is triggered. | 5 / 5 |
Progressive Disclosure | SKILL.md is a true overview: commands, flags, the subprocess contract, and cross-cutting jq/schema guidance stay inline, while per-domain detail is split into 8 one-level-deep reference files — all verified to exist and be substantive (38–170 lines each) — surfaced through a topic-index table that describes each file's coverage. This matches the 5 anchor (clear overview with well-signaled one-level-deep references); the cross-skill pointers (../lark-shared, ../lark-im) are see-also navigation to sibling skills, not nested content within this bundle. | 5 / 5 |
Total | 20 / 20 Passed |