Content
86%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 tight, highly actionable policy document: it gives exact handles, a copy-paste dispatch prompt, explicit preconditions, and a fully specified quota-failover procedure. Its only weaknesses are minor wordiness and the absence of explicit outcome-verification checkpoints.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense, purposeful prose that explains only non-obvious constraints ("one agent's `git stash` or `checkout` wipes the other's work") rather than concepts Claude already knows; a few long compound sentences and filler like "in whatever words or language" could be trimmed. | 4 / 5 |
Actionability | Fully concrete for an instruction skill: copy-paste-ready dispatch prompt (the worktree blockquote), exact handles (`instance`, `node_id`, `{{ <node_id>.output_path }}`), a branch-name example, and specific graph actions (`replan`, not `continue`) covering the common cases of new task, follow-up, parallel work, and quota takeover. | 5 / 5 |
Workflow Clarity | Rules are clearly sequenced with explicit gates ("Send a follow-up only once the instance's running dispatch has returned"; "ask the user before dispatching" when uncommitted changes matter) and a complete error-recovery loop for quota exhaustion, but there is no checklist-style verification of outcomes beyond instructing the agent to report them. | 4 / 5 |
Progressive Disclosure | A single self-contained ~75-line policy file with numbered sections and working cross-references ("section 3", "section 5"); no external references exist, and none are needed — the structure makes navigation trivial. | 5 / 5 |
Total | 18 / 20 Passed |