Content
88%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 exemplary instruction-only skill: fully executable commands, explicit sequencing with validation and error-recovery feedback loops, and disciplined imperative prose. Its only weaknesses are a triple-stated behavioral warning that could be consolidated and a length that suggests extracting a few situational sections into reference files.
Suggestions
Consolidate the three repetitions of the "a single CLI call is not completion / do not stop after implement new" warning into one prominently placed statement.
Move the orchestrator-mode section ("When an orchestrator starts this skill") into a separate reference file (e.g. references/orchestrated.md) and keep a one-line pointer plus the changed-command summary in SKILL.md.
Consider moving the detailed dependency-override and uncommitted-changes dialog trees into a reference file to shorten the main workflow body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean, imperative, and assumes Claude's competence — no basic concepts are explained — but the "do not stop after implement new" warning is repeated three times (opening blockquote, "This is a loop. Do not stop after the first step", "Concretely: do not stop after `implement new`"), which is trimmable redundancy. | 4 / 5 |
Actionability | Every instruction is copy-paste executable: exact commands with full JSON payloads (e.g. `spektacular implement new --data '{"name": "<spec_name>", "override_dependencies": true}'`), concrete refusal codes (`dependencies_unmet`, `task_not_found`, `auto_commit_failed`, `worktree_unavailable`) each paired with its exact remediation command and user-handling script. | 5 / 5 |
Workflow Clarity | The core loop (read `instruction` → perform step → `goto` → repeat until finished) is explicitly sequenced with feedback at every step via the CLI's state machine, plus detailed validation/checkpoint handling for interruptions, resume reports, uncommitted changes, merge conflicts, and the user-approved spec amendment loop with error recovery paths. | 5 / 5 |
Progressive Disclosure | No bundle files exist, and the single self-contained body is well organized under clear headers (# Worktrees, # When the spec or a design is wrong, # When an orchestrator starts this skill). It is a ~175-line body, however, and sections such as the orchestrator-mode and dependency-override details are reasonable candidates for extraction into reference files, leaving minor organization gaps. | 4 / 5 |
Total | 18 / 20 Passed |