Your to-do list has two items.
Everything else is Flow's list. A coding agent takes the task, builds it on a branch and opens a PR. GitHub Actions gates it, then QA, code-review and security reviewers read it, each in a fresh session.
Set up Flow: getflow.now/start
Tell your agent this line. Open source, Apache‑2.0.
1You ask.
In plain words, or open a GitHub issue and let Triage draft the task. Approving the task is your first approval.
2You start it.optional
Say it in a fresh session, or let Flow start a Claude Code routine on it.
3You wait.nothing to do
If the agent hits a decision that is yours, it stops, marks the task blocked and writes down the question. A failure goes back to the agent, not to you.
- flow-gates/gate build, lint, test, coveragepassed
- flow-gates/source-roots-plan scope check, flow-doctorpassed
- flow-gates/source-root already covered by the gateskipped
- flow-review/qa starts once the gate passespassed
- flow-review/code-reviewpassed
- flow-review/securitypassed
- flow-status/sync-status task moved to in reviewpassed
All 8 acceptance criteria have a proving test that asserts the actual outcome, not just that nothing threw. I re-ran the suite with no ambient environment variables: 65/65 pass.
The precedence is implemented as specified, and the ceiling applies to all three sources. A bad value fails loudly, naming the key.
No High or Critical findings. I checked that a PR can't raise its own limit: the reviewer reads the base branch's config, not the PR's.
4You merge.
Read the PR. Every criterion is ticked next to the test that proves it. That is your second approval.