agent-dispatch

# Agent Dispatch **Explicit entry point.** Invoke this by name in a harness without Claude Code commands. `/agent` carries the same routing and names the engines directly, so it does not depend on this skill and the model never loads it on its own. The router behind `/agent`. Turns a subcommand into the right action and delegates. Contains no logic of its own — delegates to `agent-architecture-audit`, `agent-config-audit`, `agent-folder-init`, and `setup-agent-routing`. ## Composition Boundary Run only the selected mode. Pass the user's target, authorized actions, and report-only restrictions to the engine. Existing explicit approval satisfies that engine's gate for the same scope; obtain approval for missing or expanded authority. Delegation never grants new host, provider, cost, publication, or production permissions. An empty or advisory mode starts no mutating workflow. ## Contract Inputs: - A single argument string (may be empty) parsed into a `mode`. Outputs: - For `audit`: a severity-ranked findings report diagnosing LLM wrapper and agent failures, with a layer-by-layer fix plan. - For `config`: an audit report covering AGENTS.override.md, AGENTS.md, configured fallbacks, CLAUDE.md, .cursorrules, hooks, and settings, with proposed fixes. - For `init`: a scaffolded `.agents/` folder structure plus root agent entry files, with a summary of files created versus skipped. - For `route`: a drafted `## Agent skills` routing block in CLAUDE.md/AGENTS.md and seeded `docs/agents/` reference files. - For _(empty)_: a one-line domain status summary and the Usage block. Creates/Modifies: - Nothing directly. The delegated skill performs any mutation behind its own confirmation gate. External Side Effects: - None by default. The delegated skills may read git remotes, `gh` APIs, or workspace files when resolving routing state. Confirmation Required: - This skill is explicit-invoke only (`disable-model-invocation`). Each delegated skill owns its own confirmation gate before writing files or changing config. This router does not relax them. Delegates To: - `agent-architecture-audit` for `audit` (diagnose LLM and agent wrapper failures). - `agent-config-audit` for `config` (audit and sync agent config files). - `agent-folder-init` for `init` (scaffold or repair the `.agents/` folder). - `setup-agent-routing` for `route` (write the dev-loop routing block). ## Step 1 — Parse the Subcommand Resolve the raw argument into a `mode`. | Argument | Mode | Delegates to | |---|---|---| | _(empty)_ | `status` | none — print a one-line domain summary + Usage block | | `audit` | `audit` | `agent-architecture-audit` | | `config` | `config` | `agent-config-audit` | | `init` | `init` | `agent-folder-init` | | `route` | `route` | `setup-agent-routing` | If the argument matches none of these, report the unrecognized input and print the Usage block — do not guess a mode. ## Step 2 — Route - **status →** print a one-line summary of the agent domain (no mutation), then show the Usage block. - **audit →** apply the `agent-architecture-audit` skill. - **config →** apply the `agent-config-audit` skill. - **init →** apply the `agent-folder-init` skill. - **route →** apply the `setup-agent-routing` skill. ## Usage ```bash /agent # status: one-line domain summary + usage /agent audit # diagnose LLM wrapper regressions, prompt/memory contamination, tool discipline failures /agent config # audit and sync AGENTS.md, overrides, fallbacks, CLAUDE.md, hooks, and settings /agent init # scaffold or repair the .agents/ folder and root agent entry files for a repo /agent route # write the ## Agent skills routing block in CLAUDE.md/AGENTS.md + docs/agents/ ``` ## Anti-Patterns - **Re-implementing logic here.** All domain logic belongs in the routed skill. - **Guessing on an unknown argument.** Print Usage instead — a wrong guess could overwrite config or scaffold into the wrong directory. - **Skipping the delegated skill's confirmation gate.** Each routed skill controls its own writes. - **Auto-running a mutating sub-skill on an empty argument.** The default mode prints status only; it never initiates a write.