The maintainer loop helps an agent choose the next bounded repository milestone from local evidence instead of waiting for a freshly written prompt after every release.
- Start from a clean synchronized
main. - Read
AGENTS.md,ROADMAP.md,CHANGELOG.md, andwork-ledger.md. - Run the local validation suite.
- Ask
./scripts/run-nextto select the next bounded action with an explicit permission flag. - Review the runner evidence in
runs/skill-runs.md. - Stop at any approval boundary before implementation, release publication, or scope expansion.
./scripts/run-next --allow evidence-harnessThis command is expected to validate the pack, identify the recommended evidence-harness milestone, write local run evidence, and stop before implementing that milestone.
Before releasing maintainer-loop changes, run:
node scripts/validate-pack.mjs .
node scripts/test-pack.mjs
node scripts/validate-maintainer-loop.mjs .
node --test
jq empty $(find . -name '*.json' -type f | sort)
git diff --checkStop and request human approval before:
- adding or changing skills
- creating real project adapters
- touching real project repositories
- publishing release artifacts
- changing credential handling
- adding package dependencies
- weakening safety, evidence, completion, or adapter rules
The loop is a decision aid and evidence writer, not a general automation daemon.