rt runs run-start
rt › runs › run-start
Pipeline: open a run DB and print its runId and runDb
Usage
bash
rt runs run-start [flags]
Arguments & flags
| Flag / Arg | Type | Default | Description |
|---|---|---|---|
--repo | text | Run-dir key for the repo | |
--work-type | text | feature | fix | ... | |
--pipeline | text | Pipeline name from the manifest | |
--run-id | text | Omit to mint one | |
--spawned-by | text | Surface that spawned this run | |
--pack-dirs | text | Colon-separated pack checkouts for provenance | |
--ticket | text | Recorded under producer work | |
--mattstack-sha | text | Appended to pack_commits | |
--mattstack-dirty | text | 1 forces pack_dirty | |
--pack-sha | text | Appended verbatim to pack_commits |
See code: commands/runs-write.ts › runsRunStart
Examples & notes
run-start is the one write verb that does not read RT_RUN_DB: it creates the run DB and hands you the path. Every other write verb reads that path from the environment, so capture it from the JSON result and export it before calling anything else:
bash
result=$(rt runs run-start --repo acme-dev --work-type feature --pipeline feature)
export RT_RUN_DB=$(echo "$result" | jq -r .runDb)
From here, rt runs stage-start, field, decision, and the rest all read $RT_RUN_DB to find the open run. When it is unset (each agent Bash call is a fresh shell), they fall back to the running run this Claude Code session recorded, else the newest running run whose worktree holds the cwd, and the JSON envelope gains runDbResolved (session or worktree). With no match they exit 2.