bootstrap
Interactive agent-flow setup for a repository. Scans the repo read-only, proposes AGENTS.md (root + per module), DOCS_INDEX.md and CONTEXT_MANIFEST.json with a confidence marker on every claim, calibrates protected paths and risk boundaries with the user, and writes each file only after the human approves it. Use when setting up agent-flow, when context files are missing, or when migrating from Root_AGENT.md. Also use whenever the user wants an AGENTS.md (or CLAUDE.md/GEMINI.md) written for a repo that doesn't have one, says their coding agents keep getting confused about the codebase, asks to "onboard" or "document" a repo for AI agents, or wants to set protected paths / risk boundaries — even if they don't say "agent-flow" or "bootstrap" by name.
gardener
Maintenance agent for agent-flow. Keeps context files true — runs /doctor, /sync-context, /audit-risk, /repair-docs and /garden, re-verifies context claims against code before refreshing timestamps, and turns recurring agent mistakes into mechanical checks. Use on a schedule, after merges that were flagged [CONTEXT_STALE], or when the user runs any of those commands. Also use whenever the user asks to check if AGENTS.md is stale or out of date, wants a health check on their context files, mentions a renamed/deleted file that context docs still point to, or wants to review the risk-audit baseline — even without naming a specific command.
implementer
Implements one issue inside its own git worktree (.worktrees/issue-N on agent/issue-N), makes the smallest change that satisfies the acceptance criteria, self-checks with the repo's own test/lint/typecheck commands, commits, and prints a JSON report for the orchestrator. Use when the orchestrator launches you with AGENT_FLOW_ROLE=implementer, or the user asks you to implement a specific issue in a worktree.
invoking-agents
Orchestrator for agent-flow. Runs one GitHub (or local) issue through Implementer → Reviewer → QA → PR as separate processes with artifact-only handoff, mechanical risk classification, a hard review-round cap, and escalation to Needs Me. Use when the user runs /implement <issue> or asks to push an issue through the pipeline. Also use whenever the user wants an issue implemented, reviewed and QA'd end-to-end with a PR at the end and a human kept out of the loop for anything low-risk — e.g. "have an agent take issue #42", "run this through implement/review/QA and open a PR", "auto-fix this bug and send a PR" — even if they don't say "orchestrator" or "agent-flow."
qa
Quality gate for agent-flow. Runs the repo's own test, typecheck and lint commands in the issue worktree, re-runs failures once to separate flaky tests from real failures, and reports raw output verbatim as JSON. Never modifies code. Use when the orchestrator launches you with AGENT_FLOW_ROLE=qa after a review is approved.
reviewer
Read-only code reviewer for agent-flow. Reviews one diff against the issue's acceptance criteria, repo rules in AGENTS.md, protected paths and risk boundaries, and returns a JSON verdict (approved / request_changes) with categorized, line-anchored findings. Use when the orchestrator launches you with AGENT_FLOW_ROLE=reviewer or the user asks for an agent-flow review of a diff.