Content
92%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 operational overlay: it stays compact while packing in exact commands, tool-call shapes, error codes, and validation gates, and it delegates the bulk workflow to real, well-organized one-level-deep reference files. Its only weakness is a handful of overlong run-on rules that could be tightened. Scores are strong across all dimensions.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, rule-driven, and teaches only environment-specific facts Claude could not know (CLI commands, tool surface, error codes) — no padding or generic concept explanations. A few passages could be trimmed, e.g. the ~70-word run-on Non-Negotiable about goal registration and the git tree-binding parenthetical, so it matches anchor 4 (efficient, minor over-explanation) rather than anchor 5. | 4 / 5 |
Actionability | Guidance is copy-paste ready throughout: exact CLI invocations ("omo-agent-toolkit ulw-loop create-goals", "--session-id <id>", "git rev-parse --short \"HEAD^{tree}\""), full tool-call shapes ("spawn_agent({\"task_name\":\"<lower_snake_id>\"...\"fork_turns\":\"none\"})"), and concrete error-code contracts (ULW_LOOP_SESSION_SCOPE_REQUIRED, ULW_LOOP_CODEX_GOAL_JSON_INVALID). The tool-mapping table covers the common cases (spawn, re-task, send, wait, enumerate, stop), matching anchor 5. | 5 / 5 |
Workflow Clarity | "Required First Steps" gives an ordered bootstrap sequence with an explicit precondition ("before running any ULW command or recording evidence"), validation gates are pervasive ("Record only after cleanup receipts exist", tree-hash evidence binding with a re-run rule, "lands when its own evidence is captured and its gates are green"), and feedback loops exist for error recovery (fallback/respawn rules, lead-resolved conflicts). Despite batch delegation and merge operations, validation is explicit, so the destructive/batch cap does not apply — this matches anchor 5. | 5 / 5 |
Progressive Disclosure | The bundle structure is exactly as signaled: SKILL.md states "The full workflow lives in references/full-workflow.md. Read only the sections needed for the current phase", and both referenced files (full-workflow.md, 272 lines; define-goal.md, 107 lines) exist with the named sections (Bootstrap, Execution Loop, Manual-QA channels, Stop Rules) verified present. References are one level deep (the only cross-reference is the sibling define-goal.md), the split is appropriate (bulk detail like quality-gate JSON shapes lives in the reference), matching anchor 5. | 5 / 5 |
Total | 19 / 20 Passed |