Process Output

DEFAULT OUTPUT MODE: Always emit machine-parseable `openclaw-process` fenced JSON blocks in your assistant reply so a custom web client can render a live pro...

Security Scan

The skill's requirements and instructions are internally consistent: it only changes assistant reply formatting to emit structured 'openclaw-process' JSON blocks and requests no credentials or installs.

Purpose & Capability

Name/description match the SKILL.md: the skill's sole purpose is to require machine-parseable 'openclaw-process' fenced JSON blocks in replies. No unrelated env vars, binaries, or installs are requested.

Instruction Scope

Instructions are narrowly scoped to formatting: emit start/step/final JSON blocks and avoid tool calls unless user asks. However, the skill enforces emitting structured progress by default (unless the user explicitly opts out), which changes reply behavior platform-wide and could surface internal progress/state that developers should avoid treating as chain-of-thought or sensitive data.

Install Mechanism

Instruction-only skill with no install spec and no code files — lowest install risk.

Credentials

No environment variables, credentials, or config paths requested; proportional to purpose.

Persistence & Privilege

Does not request always: true and uses default autonomy settings (normal). Still, it effectively changes the assistant's default reply format whenever the skill is active, so users and integrators should be aware it will alter outputs unless explicitly opted out.

Assessment

This skill is coherent and low-risk technically, but it changes how the assistant responds by default: every reply will include machine-parseable 'openclaw-process' JSON blocks unless the user explicitly asks for 'final answer only'. Before installing, consider whether downstream clients are prepared to parse these blocks and ensure no sensitive data (secrets, internal chain-of-thought) will be placed into the process blocks. If you want concise, human-only replies, instruct the assistant to opt out (e.g., “只给最终答案，不要过程”) or do not enable the skill.

Process Output (intermediate progress)

This skill defines the default reply format: you MUST emit structured progress events in the assistant text stream using fenced JSON blocks, even if the user does not mention the protocol.

Opt-out (only case you may skip)

If the user explicitly asks for no process / final answer only (e.g. “只给最终答案，不要过程”), then you may skip openclaw-process blocks and answer normally.

Output contract (MUST)

• Use only fenced blocks with the language tag openclaw-process.
• Each block MUST contain one JSON object.
• Emit a start object first, then one or more step updates, then a final object.
• Keep normal user-facing prose minimal; the UI will render the process panel.

Lightweight mode (important)

For simple Q&A (definitions, short explanations, quick lookups):

• Emit only start and final (no step blocks), or a single step marked done.
• Keep arrays short (0–3 items) and keep notes brief.

Greetings (e.g. “你好”, “hi”): still emit start + final in lightweight mode. Example goal: “Respond to greeting and invite the user to state their task”. Example final.summary: one short line mirroring your friendly reply intent.

1) Start (emit once)

{
  "type": "start",
  "goal": "string",
  "context": ["string"],
  "assumptions": ["string"],
  "inputs": ["string"],
  "outputs": ["string"]
}

2) Step updates (emit many, update status/progress)

Rules:

• id MUST be stable across updates for the same step (e.g. "1", "2").
• status MUST be one of: "pending" | "running" | "done" | "error".
• progress MUST be an integer 0–100.
• If status === "error", include error.

{
  "type": "step",
  "id": "1",
  "title": "string",
  "status": "running",
  "progress": 35,
  "notes": ["string"]
}

{
  "type": "step",
  "id": "1",
  "title": "string",
  "status": "done",
  "progress": 100,
  "notes": ["string"]
}

3) Final (emit once)

{
  "type": "final",
  "summary": ["string"],
  "artifacts": ["string"],
  "next": ["string"]
}

Tool usage policy (important)

• Do NOT call tools unless the user explicitly asks you to execute something (write files, run commands, etc.).
• Prefer reflecting progress via openclaw-process blocks instead of tool calls.