Content
77%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 exceptionally actionable, well-sequenced orchestration skill with strong validation gates for its destructive and irreversible operations. Its weaknesses are token redundancy — the core spawn/worktree facts are restated several times — and the absence of any progressive disclosure: a 285-line monolith where the prompt template, worked example, and pruning guide could live in reference files.
Suggestions
State each core fact once: the 'spawn_task forks a fresh throwaway worktree' behavior, the group-placement limitation, and the never-`git add -A` rule are each repeated 2-4 times across the intro, steps, output text, and Important section — consolidate to one authoritative statement and cross-reference it.
Move the full prompt template and/or the worked MES-4282 example into references/ (e.g. references/spawn-prompt-template.md) and keep only the field list and one-line rationale in SKILL.md, cutting main-file tokens substantially.
Relocate the orphan-worktree pruning instructions to a reference file and keep a one-line pointer in Important, since pruning is periodic maintenance rather than part of the handoff path itself.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most content is genuinely non-obvious operational fact Claude cannot know (spawn_task worktree semantics, husky bypass, worktree/branch object-DB mechanics), so it earns its tokens — but key facts are repeated three to four times ('spawn_task ALWAYS forks a fresh throwaway worktree' appears in the intro, step 9's rationale, step 11's output text, and the Important section; group-placement and never-`git add -A` are each restated). Mostly efficient but could be tightened by roughly a third by stating each fact once and cross-referencing. Not a 2 because nearly every passage carries information a cold agent could not reconstruct. | 3 / 5 |
Actionability | Fully executable throughout: exact bash snippets for branch detection, artefact staging, MR lookup, and orphan pruning; lookup tables for chapter titles and invocation forms; a complete fill-in prompt template plus a fully worked MES-4282 example; and verbatim output strings for every branch of the flow. This matches the copy-paste-ready anchor covering the common cases. | 5 / 5 |
Workflow Clarity | Eleven clearly numbered steps with explicit validation checkpoints and error-recovery loops: STOP if on the default branch (step 1), distinguishing a genuine commit no-op from failure with a loud WARNING instead of silent success (step 4b), a user-facing AskUserQuestion gate before the irreversible spawn (step 10), and safe orphan-pruning guards ('refuses if dirty', 'only if `git log <default>..<branch>` is empty'). This matches the anchor for explicit validation steps and feedback loops. | 5 / 5 |
Progressive Disclosure | Structure is good (## Steps, ## Important, tables, clearly labeled templates) and references none, but nothing is offloaded: the ~75-line prompt template plus worked example and the orphan-worktree pruning guide are prime candidates for a references/ file, and at ~285 lines the single SKILL.md inlines content that clearly belongs in separate files. No bundle directories exist, so all content is inline — matching the 'structure present but content that should be separate is inline' anchor rather than the well-split anchor at 4. | 3 / 5 |
Total | 16 / 20 Passed |