normies-workflow

name: normies-workflow description: Orchestrate Docker-isolated, branch-based multi-agent git workflows with normies. Use for parallel edits, batched tasks, retries, and explicit review/integration gates. Prefer this over ad-hoc shell loops for multi-step coordination; skip for trivial one-file edits that do not need orchestration. compatibility: Requires git, docker daemon, and normies CLI in PATH.

Normies Workflow

When To Use

• Requests include multi-agent orchestration, parallel edits, batched fixes, or branch-isolated tasks.
• Work needs retries (retry --failed) or explicit local review/integration gates.
• A repository workflow should be deterministic and auditable from run artifacts.

When Not To Use

• A trivial single-file edit that does not need orchestration.
• A one-off shell command where run metadata and branch isolation add no value.

Run Workflow

1. Confirm prerequisites. Check normies --help. Run normies doctor --repo <repo>.

2. Build a spec. Use normies init for a guided wizard (minimal or baseline template). Use normies init --agent-context codex,claude to scaffold AGENTS/Claude guidance. Use --dry-run to preview file writes and --force to replace unmanaged Claude skill files. Use normies make-spec for fast, valid JSON specs. Read references/spec-guide.md for full field details and patterns.

3. Execute the pipeline. Run normies run --repo <repo> --spec <spec> --jobs <N>. Run normies retry --run-id <run_id> --failed --jobs <N> if any agents failed. Run normies review --latest (or --run-id <run_id>). Run normies integrate --latest (or --run-id <run_id>). Use the printed push command from integrate output to push manually.

4. Triage failures. Run normies status --latest. Run normies logs --latest --list-agents. Run normies logs --latest --agent <agent_name> --tail 200. Run normies logs --latest --agent <agent_name> --follow for live logs. Inspect .orchestrator/runs/<run_id>/run.json for full per-agent state.

5. Clean up worktrees after completion. Run normies cleanup --latest. Add --remove-run-dir when you also want to remove run metadata and artifacts.

Execution Defaults

• Keep needs_network disabled unless the task requires network.
• Keep auto_commit enabled unless explicit manual commit behavior is needed.
• Keep per-agent commands idempotent and non-interactive.
• Keep review checks explicit in review.required_checks.
• Use --json when output is consumed by scripts/automation.

Status Interpretation

• Treat failed as execution failure (container/command failed).
• Treat blocked as unresolved workflow state (for example dirty worktree with auto_commit: false).
• Treat no_change as valid no-op.
• Treat ok as successful output ready for review/integration.

Resources

• references/spec-guide.md: Review spec schema, command sequence, and troubleshooting tips.