worklisted

# work -- the do-a-unit-of-work orchestrator You run one unit of work through the pipeline compiled into this skill. Everything below the stage list is baked: you never resolve a stage, a binding, or a chain -- the compiler already did. ## 1. Work type {{work-type}} ## 2. Stages Read the list for the chosen work type. Each entry is a compiled stage skill sitting beside this one; `dir` is where to read it. {{pipeline.stages}} ## 3. Start the run ```bash PACK_DIRS="$(cd "${CLAUDE_SKILL_DIR}/../.." && pwd -P)" ``` `PACK_DIRS` is the pack root by the compiler's layout; `run-start` records the pack's commit and dirty state only when that directory is a git checkout (an installed plugin cache is not), so absent provenance there is expected. Then run `run-start` with the flags for the chosen work type, adding `--ticket <id>` when the request named one and `--spawned-by "<surface>"` when this run was spawned rather than started interactively. Never fabricate a ticket. {{run-start.flags}} The response must parse as JSON with `ok: true` and a `runDb`. Anything else (a listing of runs, usage text) means this rt predates the run DB write verbs: stop, and tell the user to update rt before continuing. Do not proceed without a `runDb`. ```bash rt runs run-start <flags for the work type> --pack-dirs "$PACK_DIRS" [--ticket <id>] [--spawned-by "<surface>"] export RT_RUN_DB=<runDb from the response> # each tool call is a fresh shell: prefix every rt runs command with RT_RUN_DB=<r