Content
60%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 delivers dense, highly actionable operational guidance with concrete commands, tool-call examples, and explicit validation rules, and it points cleanly to real one-level-deep references. Its main weakness is token efficiency: harness-routing rules are stated two to three times over, including one near-unreadable wall-of-text paragraph, which inflates the skill's context cost without adding information.
Suggestions
Consolidate the duplicated harness-routing guidance: state the V1/V2 tool-surface rules and the "this section wins" precedence rule once, and delete their repeats in the Non-Negotiables and Codex Tool Mapping sections.
Break the line-24 wall-of-text paragraph into a short list or move it, along with the OpenCode-to-Codex translation table, into a dedicated reference file so the SKILL.md body stays a compact overview.
Merge the repeated wait/backoff and WORKING/BLOCKED protocol bullets (lines 30 and 56-57) into a single subagent-supervision rule.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body duplicates the same guidance repeatedly: "If a code block below conflicts with this section, this section wins" appears three times, the wait/backoff rule and WORKING/BLOCKED protocol each appear twice, and V1/V2 tool-surface routing is explained in two separate sections (the opening compatibility section and the later Codex Tool Mapping). Line 24 is a dense wall-of-text paragraph. This is noticeably verbose with several padded, redundant passages rather than merely "could be tightened". | 2 / 5 |
Actionability | Guidance is largely executable: exact CLI commands ("omo-agent-toolkit ulw-loop create-goals", "git rev-parse --short \"HEAD^{tree}\""), copy-paste spawn-call JSON in the mapping tables, concrete error codes (ULW_LOOP_SESSION_SCOPE_REQUIRED), and a two-condition team-mode decision rule. Minor gaps remain because key execution detail (delegation specifics, per-criterion cycle) is deferred to references without inline summaries. | 4 / 5 |
Workflow Clarity | "Required First Steps" gives a clear numbered entry sequence naming the exact reference sections to read before acting, and validation is explicit (evidence bound to a git tree, gated checkpoint, stop rules, resume-after-compaction rule). Some checkpoints live only in the referenced workflow rather than the body, leaving minor validation gaps relative to the anchor-5 example. | 4 / 5 |
Progressive Disclosure | The bundle structure is good: two real, one-level-deep reference files (full-workflow.md, define-goal.md), clearly signaled with phase-specific reading instructions ("Read only the sections needed for the current phase"). However, roughly 30 lines of harness-compatibility translation tables are inlined in the body and duplicated across two sections — content that belongs in a reference file — so it falls short of the clean overview anchor 5. | 4 / 5 |
Total | 14 / 20 Passed |