agentflow（agentflow 搜索）

Agentflow

Agentflow is a supervised local runtime for long-running coding work. Humans author a graph with intent and outcome boundaries; Codex CLI or Cursor CLI executes substantial nodes; the supervisor records bounded interventions; terminal runs produce a delivery package.

Route By Task

• Author or review a graph: read references/graph-authoring.md.
• Need exact fields: read references/graph-contract.md.
• Choose managed patterns: read references/managed-workflows.md.
• Need CLI validation or launch behavior: read references/cli-and-validation.md.
• Debug failures, resume, or inspect delivery: read references/run-debugging.md.
• Need failure semantics: read references/failure-and-validation.md.
• Need examples: read references/examples.md.
• Need reusable plugin workflows or tools: use agentflow-plugins.

Default Workflow

1. Confirm the graph has intent.goal, intent.acceptance_criteria, explicit constraints, explicit repos, and explicit profiles.
2. Prefer fewer, larger outcome nodes with named artifacts and node-level goal plus acceptance_criteria.
3. Set supervision budgets and delivery sections appropriate to the task.
4. Use plugin-bundled CLI tools for team capabilities; verify each tool's capability, impact, and credential requirements.
5. Run agentflow plugin resolve --graph <path> when plugins are declared.
6. Run agentflow validate --graph <path>.
7. Run agentflow validate --graph <path> --run-ready before launch on this machine.
8. Run agentflow validate --graph <path> --show-compiled for managed patterns, plugin workflows, repeat scopes, or nontrivial artifact handoffs.
9. After a run, inspect delivery/reviewer-guide.md, delivery/manifest.json, delivery/run-map.md, and declared artifacts before raw runtime files.

Authoring Posture

• Treat the authored DAG as the human contract, not a prose plan.
• Use context for node material and artifacts for durable handoffs.
• Treat repos and profiles as operational authority; put scope boundaries and out-of-scope notes in constraints.
• Keep downstream references on named artifacts from public node ids.
• Use deterministic checks for hard facts and AI checks for semantic judgment.
• Make high-impact limits explicit in constraints before granting external, secret, or mutation tools.
• Do not widen scope through supervisor behavior; use checkpoints or graph edits for human decisions.

Runtime CLI Posture

• Humans use agentflow; agents inside running nodes use af.
• af is injected into agent nodes on PATH and reads $AGENTFLOW_RUNTIME_METADATA.
• Use af --help and af <command> --help for exact runtime CLI arguments, defaults, output shape, examples, and safety notes.
• Prefer af status, af tools list, and af context show when debugging what a node actually received.
• Prefer af artifact write for declared handoffs instead of ad hoc output files.
• Use af log --type for worker evidence and helper coordination notes, but keep durable conclusions in artifacts.
• Treat af spawn helpers as supervised sessions with their own artifacts, not persistent coworkers.