Content
81%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 strong operational skill: every step is executable with exact commands and message templates, the sequence is clear with explicit validation (marker check, result review) and recovery paths for failure and partial completion, and the single reference is well signaled. The weaknesses are localized: a dense code-internals paragraph in Step 2b and redundant prompt-crafting guidance spread across three sections.
Suggestions
Move the run-coding-cli.sh enforcement internals (exit 125, processor/run-id ownership, legacy marker rules) to references/delegation-guidance.md, keeping only the actionable instruction: check the marker, wait if present, never reuse another operation's run ID.
Consolidate 'Tips for Better Prompts' into Step 3 so prompt-crafting guidance lives in one place and the overlap with 'Signs of a weak prompt' is removed.
Add a brief note on how to wait/retry when the .processing marker exists (e.g. poll interval or where completion is announced), since the current guidance just says 'wait and retry'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient, command-driven prose with no filler explaining what git or CLIs are, but two areas could be tightened: the Step 2b paragraph on code internals ("run-coding-cli.sh exits 125... Active legacy markers without run IDs block execution; expired markers are ignored... does not replace atomic scheduling or provide a permission boundary") is dense reference-level detail, and prompt-crafting guidance is spread across Step 3, the worked example, and 'Tips for Better Prompts' with overlap ('Be surgical: change line 47 from X to Y' restates 'Signs of a weak prompt'). Not a 2 because padding is localized; not a 4 because the 2b paragraph exceeds a minor instance. | 3 / 5 |
Actionability | Fully executable throughout: copy-paste bash blocks (mkdir, mc mirror, the .processing marker test), exact message templates for coding-request / coding-result / coding-failed / completed, and a completely specified example prompt with file path, line numbers, JSON response shapes, and a scope constraint ('Do not change any other files'). Success, failure, and follow-up-round paths are all covered concretely. | 5 / 5 |
Workflow Clarity | The workflow is explicitly sequenced (prepare workspace → push to MinIO → check marker → craft prompt → send → wait → 6a/6b branches) with a pre-flight validation checkpoint (marker check with wait-and-retry), post-run validation (agentteams-sync, git diff, 'Run tests if applicable', review checklist), and genuine feedback loops (coding-failed → manual fallback and report; 'Multiple Rounds' for partially completed work). Checkpoints and recovery are explicit, matching the level-5 anchor. | 5 / 5 |
Progressive Disclosure | Good structure with clear section headers and one well-signaled, one-level-deep reference — 'Operational reference: references/delegation-guidance.md — one-shot delegation from the delegator's side: what it is, task-spec discipline...' — and the referenced file exists with content matching that scope. Not a 5 because the marker-enforcement internals in Step 2b are reference-level material inlined in SKILL.md, and the prompt guidance spanning three sections could be consolidated or split into the reference. | 4 / 5 |
Total | 17 / 20 Passed |