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.
The body is an unusually dense, highly executable operational directive: exact commands, named tools, explicit validation gates, and strong feedback/stop rules make it near-copy-paste actionable. Its weaknesses are duplication across sections that inflates token cost and a monolithic single-file structure that inlines reference-grade detail (subagent routing compatibility, per-channel QA) instead of splitting it into one-level-deep bundle files.
Suggestions
Deduplicate repeated rules — state the never-clear-browser-data, tmux capture-pane, and fork_context guidance once and cross-reference it — to cut token cost without losing content.
Move the Codex subagent routing/TOML compatibility sections and the per-channel QA command details into reference files (e.g. references/subagent-routing.md, references/qa-channels.md) linked one level deep from the body.
Compress environment-specific tool-surface notes (multi_agent_v1 vs v2 differences) into a short table, since much of it restates what the spawn-tool schema itself exposes.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Sentences are telegraphic and never re-explain concepts Claude already knows, but there is real duplication that could be tightened: the never-clear-cookies/caches warning appears in both the browser channel and Cleanup, the tmux capture-pane prohibition appears twice, and fork_context guidance is restated across three sections. Fits anchor 3 (mostly efficient but could be tightened) rather than 4, since the duplication is more than minor. | 3 / 5 |
Actionability | Fully executable throughout: literal commands ("NOTE=$(mktemp -t ulw-$(date +%Y%m%d-%H%M%S).XXXXXX.md)", "node script/qa/web-terminal-visual-qa.mjs --title ... --command ...", "git log --oneline -20", "tmux kill-session -t ulw-qa-<criterion>"), exact tool names (create_goal, update_plan, multi_agent_v1.spawn_agent), exact env vars, and a mandated literal first line. Specific examples cover the common cases. | 5 / 5 |
Workflow Clarity | Clear sequence with explicit validation and feedback loops: an ordered bootstrap (survey → goal → notepad → todos), an 8-step execution loop with blocking checkpoints (LSP diagnostics clean, read-back teardown receipts, rerun-what-moved), a reviewer loop with delta re-reviews, a 2-attempt stop rule, and an explicit stop-goal checklist including cleanup-state verification. | 5 / 5 |
Progressive Disclosure | Internal section headers are clear, but the body is a monolithic ~490-line single file with no bundle references at all — the Codex subagent TOML-routing compatibility details and channel-specific QA commands (150+ lines) clearly belong in separate reference files per the one-level-deep model. Matches anchor 3 (some structure, content that should be separate is inline). | 3 / 5 |
Total | 16 / 20 Passed |