Conciseness

The body is dense and operational — tables, commands, and routing rules with almost no explanation of concepts Claude already knows. Minor redundancy could be trimmed: the four agent names are enumerated three times (Quick Dispatch rule, Agent Catalog, Error Handling), and the "When to Use Which Agent" / "User Intent Routing" tables overlap somewhat. Not a 5 because of that repeated enumeration; clearly above anchor 3 since nothing is padded or over-explained.

Actionability

Fully executable, copy-paste-ready commands cover the common cases: `omo-agent <agent> "<prompt>"`, `--model <alias>`, `--file`/`--skill` attachment, explicit Bash timeout of 600000, run_in_background for parallel dispatch, and a template prompt with OUTPUT instructions. Intent-routing and model-alias tables make the right invocation unambiguous.

Workflow Clarity

The dispatch sequence is clear — recognize agent/model from arguments, invoke via omo-agent with the right execution mode, collect all results, then synthesize — and the Error Handling section provides explicit feedback loops (exit-code semantics, failure signatures mapped to remedies, "If an agent modified files, verify (tests, diagnostics)"). Not a 5: checkpoints are scattered across sections rather than a coherent ordered flow, and some rules are partial (e.g., "Never parallelize sequential deps" lists only one example).

Progressive Disclosure

The in-body structure and signaling are good — prompt-block composition is properly offloaded to the one-level-deep references/prompt-blocks.md with a quick-reference table — but most referenced paths are dead: models-routing.md, gotchas.md, and all four agents/*.md links (oracle, explore, librarian, multimodal-looker) do not exist in the bundle. Scored against the actual bundle per the guidelines, the navigation pattern is only half-implemented, matching anchor 3 (structure present but references fail to deliver).

Specificity

The description lists several concrete capabilities — "bridges Claude Code to OpenCode CLI, handles agent catalog, model routing flags, and prompt-block construction" — plus named models (GPT-5, Gemini, Codex, o3) and agents (oracle, explore, librarian, multimodal-looker). Not a 5: "handles agent catalog" is a functional area rather than a crisp action, and coverage of what the bridge actually does has minor gaps.

Completeness

Both halves are explicit: "Use when user asks for non-Claude model dispatch (GPT-5, Gemini, Codex, o3) or invokes OMO agents (oracle, explore, librarian, multimodal-looker)" is a concrete when-clause with trigger phrases, and the rest states what it does (bridging, agent catalog, model routing flags, prompt-block construction). Third-person voice is used correctly.

Trigger Term Quality

Good natural keyword coverage: users would naturally say "GPT-5", "Gemini", "Codex", "o3", or the OMO agent names, and the phrase "user asks for non-Claude model dispatch" captures the intent. A few common natural phrasings (e.g., "ask gpt", "use gemini", "multi-model", "opencode") are absent from the description itself, keeping it just below comprehensive anchor 5.

Distinctiveness Conflict Risk

Clear niche — dispatching non-Claude models via OpenCode's agent system — with distinct triggers (specific model and agent names) that no generic skill would match. Conflict risk with other skills is minimal.