route-engineering-worklisted

# Route Engineering Work ## Purpose and authority Classify one request, emit one routing decision, and stop. Remain read-only. Never plan, edit source, publish, commit, push, create a pull request, invoke a selected skill, or create an input file for conversational text. ## Start Resolve `skill-root` to this directory. Read [Routing Policy](references/routing-policy.md), then run the stateless CLI with facts already available to the caller: ```bash python /absolute/skill-root/scripts/cli.py --repo-root /absolute/repo \ --input request="<request>" --format json run ``` Use `--request-file` only when that file already exists. Supply repository, approved-plan, or issue facts only when already known; the router validates paths but does not inspect plan contents or contact GitHub. ## Next-step loop 1. Run the router once from `skill-root`. 2. Read the single decision from `result`. 3. Do not invoke a prerequisite, primary skill, or follow-up. 4. Preserve the exact allowed and forbidden actions in the decision. The direct `scripts/route_engineering_work.py` command remains the legacy JSON entry point; use the common CLI when a protocol envelope is required. ## Completion and recovery Return the exact JSON object in `result` with no prose or follow-up execution. Complete only when it contains exactly the routing-decision schema fields and the target repository is unchanged. Invalid or missing inputs fail closed without a routing decision. Correct the caller-supplied fa