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-validated workflow — concrete commands, exact tool payloads, guarded commits, explicit failure paths, and a user gate — that scores at the top of actionability and workflow clarity. Its costs are token efficiency and structure: the same three design facts are restated three to four times across the intro, steps, and Important section, and everything (worked example, pruning guide) is inlined in one long file with no reference files. Consolidating the repeated rationale and splitting the example/pruning material into references would raise both dimensions.
Suggestions
State each core fact once — e.g. consolidate the fresh-worktree/durability/group-placement rationale (currently repeated in the intro, step 4b/9, step 11's output, and Important) into a short "Key constraints" block and reference it from the steps instead of restating it.
Move the worked example prompt (step 9) and the orphan-worktree pruning guide (Important) into reference files (e.g. references/example-prompt.md, references/orphan-pruning.md) and link to them, keeping SKILL.md as a lean overview.
Trim the multi-line justification comments (e.g. the husky/--no-verify rationale and the wrapper-policy paragraph, which also appears in Important) to one-line pointers; the operational command matters more than the repeated motivation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly skill-specific knowledge Claude cannot know (spawn_task worktree behavior, devflow wrapper policy, husky bypass), so it avoids explaining known concepts — but the same core facts are restated repeatedly: the "spawned session lands in the wrong worktree" warning appears in the intro, step 9's template, the worked example, and the step-11 output; the durability/commit rationale appears in intro point 2, step 4b, and Important; group-placement limits appear three times. This fits the level-3 anchor ("mostly efficient but ... could be tightened") rather than 2, because the verbosity is redundant restatement of bespoke facts rather than padded generic explanation, and rather than 4 because the duplication is more than minor. | 3 / 5 |
Actionability | Guidance is fully executable throughout: copy-paste bash for branch/worktree detection, slug sanitization, MR detection via gh/glab, the commit-with-guardrails block; an exact `mark_chapter` payload; a printf terminal-title escape; the exact AskUserQuestion options and exact output strings to print; and a complete spawn prompt template plus a fully worked MES-4282 example. This matches the level-5 anchor (copy-paste ready, examples cover the common cases) and exceeds level 4, which allows minor gaps — the only placeholders are inherently run-specific values. | 5 / 5 |
Workflow Clarity | Steps 1–11 are explicitly sequenced with strong validation checkpoints and feedback loops: STOP if on the wrong branch before committing (step 1); a guarded no-op vs. real-failure distinction in step 4b where a failed commit prints a loud WARNING rather than passing silently; append-not-overwrite re-entry handling; the `--no-handoff` and `--unattended` escape hatches; and a user confirmation gate before the spawn. This matches the level-5 anchor (explicit validation, error recovery, escape hatches) and is above level 4, whose "minor validation gaps" are absent. | 5 / 5 |
Progressive Disclosure | There are no bundle files (no references/, scripts/, or assets/), so everything lives in one ~290-line SKILL.md. Internal structure is good (numbered steps, tables, an Important section), but content that belongs in separate files is inlined: the 35-line worked example duplicates the template, and the orphan-worktree pruning guide and wrapper-policy rationale are embedded in Important. This matches the level-3 anchor ("some structure ... content that should be separate is inline"); it is above 2 because structure is far from minimal, and below 4 because there are no well-signaled external references and the file is well over the simple-skill threshold. | 3 / 5 |
Total | 16 / 20 Passed |