Content
78%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.
This is a well-engineered instruction-only skill: compact, jargon-free of generic explanations, with concrete commands, exact tool-call signatures, and disciplined gating of evidence and checkpoints. Its only real costs are a few overloaded run-on bullets and end-to-end loop sequencing that lives in the reference rather than the body. Progressive disclosure is exemplary, with verified, clearly-signaled one-level references.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational with essentially no padding — every rule is CLI- or harness-specific knowledge Claude would not already have (session scoping, error codes like ULW_LOOP_SESSION_SCOPE_REQUIRED, tree-hash evidence binding, tool version mapping). It misses the 5 anchor because several bullets are long run-ons (e.g., the register-goals bullet interleaving four rules in one sentence) and the team-mode rationale prose could be tightened without losing the decision criteria. | 4 / 5 |
Actionability | Guidance is largely executable: exact commands (omo-agent-toolkit ulw-loop create-goals, --session-id, CODEX_THREAD_ID, git rev-parse --short "HEAD^{tree}"), concrete tool-call signatures in the mapping table (spawn_agent({"task_name":..., "fork_turns":"none"})), and message protocols (TASK:/DELIVERABLE/SCOPE/VERIFY, WORKING:, BLOCKED:). It stops short of 5 because some commands are named but not shown with full syntax (checkpoint --codex-goal-json, record-evidence, team_create), deferring exact usage to the reference. | 4 / 5 |
Workflow Clarity | The Required First Steps give a clear, explicitly gated pre-flight sequence ('before running any ULW command or recording evidence', 'Goal creation is NEVER skipped'), and the delegated workflow's phases (Bootstrap, Execution Loop, Stop Rules) are named so the sequence is traceable into the reference, which itself contains gated checkpoints ('Only after that gated checkpoint succeeds'). Not 5 because the body-level view of the execute→verify→record loop is indirect — the reader must assemble the full checkpoint cadence from the reference rather than seeing an end-to-end sequence with feedback loops inline. | 4 / 5 |
Progressive Disclosure | The body is an intentionally compact overview ('The full workflow lives in references/full-workflow.md. Read only the sections needed for the current phase') with exactly two one-level-deep references, both verified to exist, and every cited section (Bootstrap, Execution Loop, Manual-QA channels table, Stop Rules in full-workflow.md; goal shaping in define-goal.md) is actually present and clearly signaled with when to read it. Always-required rules stay inline while bulk detail is split out, matching the clear-overview anchor. | 5 / 5 |
Total | 17 / 20 Passed |