Content
70%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.
A strong, highly executable orchestration workflow with explicit validation checkpoints and intervention feedback loops for a risky batch operation. The main weaknesses are a verbatim-duplicated guidance block and heavy inline CLI reference tables that belong in a separate reference file.
Suggestions
Delete the duplicated model-selection paragraphs for OpenCode/Copilot/Goose (they repeat the bullet list verbatim) and keep only the bulleted version.
Move the provider/model constraint table, auto-approve flags table, known auto-approve issues, and model-id snapshot into a references/cli-reference.md, leaving a one-line pointer in SKILL.md.
Make Monitor and Merge executable: specify how each CLI's output is captured (e.g., redirect to worktrees/<name>/output.log) and give the concrete merge command (e.g., 'git merge farm/<name>' or 'gh pr create').
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense, high-value reference material (CLI flags, model ids), but the model-selection guidance for OpenCode/Copilot/Goose is duplicated verbatim — once as bullets ('OpenCode: prefix with provider, e.g. `anthropic/claude-sonnet-5-5`') and again as repeated plain paragraphs — and the all-caps confirmation block could be tightened. This is more than 'minor instances of over-explanation' (anchor 4) but not 'several unnecessary explanations' across the skill (anchor 2). | 3 / 5 |
Actionability | Launch commands are copy-paste ready per CLI (e.g., 'claude -p "PROMPT" --dangerously-skip-permissions --model claude-sonnet-5-5 --max-turns 30 --output-format json') along with concrete worktree setup and cleanup commands. Minor gaps keep it below 5: the merge step ('Merge worktree branches into the working branch') and the Monitor step ('read terminal output files') give direction without executable commands, and no command shows how output is captured to files. | 4 / 5 |
Workflow Clarity | An explicit 8-step sequence (Plan through Report) with concrete validation checkpoints — 'Verify each worktree is clean', pre-flight auto-approve verification, 'Validate: tests pass, no unintended changes, scope respected' — plus error-recovery feedback loops (restart on failure, reassign after '3+ retries', escalate to user). The batch/destructive-operation cap does not apply since validation steps are explicitly present; this fits anchor 5 rather than 4 because checkpoints are comprehensive, not just 'most'. | 5 / 5 |
Progressive Disclosure | Sections are well organized (role, core_concepts, process, resources), but roughly 60% of the body is inlined CLI reference material — the provider/model constraint table, auto-approve flag table, known-issues list, and model-id snapshot — that clearly belongs in a separate references/ file; no bundle files exist. Good structure with content that should be separate inline matches anchor 3; it is above anchor 2 (structure is clear, not minimal) and below anchor 4 (bulk reference content is not appropriately placed in separate files). | 3 / 5 |
Total | 15 / 20 Passed |