---

name: x-autonomy-v2

description: Official X API v2 read/write autonomy for OpenClaw agents (safe inbox-first engagement)

---

X Autonomy v2 (Safe Mode)

Use official X API v2 only (OAuth2 PKCE). No browser posting. No cookie/private GraphQL writes.

Core policy

Allowed (default)

1. Inbox autonomy (mentions + replies to our own tweets)

2. Discovery drafting (search, score, draft)

3. Optional auto-quote with strict caps

Disallowed

• Mass unsolicited auto-replies to strangers
• Keyword spam loops
• Duplicate/template replies

Guardrails

• Max 8 replies/hour
• Max 40 likes/day
• Max 20 retweets/day
• Jitter 20–120s between write actions
• One reply per user per thread unless they reply again
• Human approval required for political/legal/medical/financial/high-conflict content

Required env

• `X_CLIENT_ID`
• `X_REDIRECT_URI`

Optional:

• `X_SCOPES`
• `X_TOKEN_STORE_PATH`
• `X_ACTION_LOG_PATH`

Required scopes

• `tweet.read`
• `tweet.write`
• `users.read`
• `like.write`
• `offline.access`

Tool contract

Every write action must accept `idempotency_key` and `dry_run`.

Every write action must append to audit log (`x_actions.ndjson`).

Tools to implement

• `x_auth_status`
• `x_auth_start`
• `x_auth_callback`
• `x_home_timeline`
• `x_mentions`
• `x_thread`
• `x_search_recent`
• `x_post`
• `x_reply`
• `x_quote`
• `x_thread_post`
• `x_like`
• `x_unlike`
• `x_retweet`
• `x_unretweet`

Agent operating modes

Mode A: Inbox (default)

1. Poll mentions/replies since last_seen_id

2. Load thread context

3. Classify + draft

4. Auto-execute only if safe and confidence >= 0.75

5. Persist state + logs

Mode B: Discovery (controlled)

1. Search query packs

2. Relevance/risk scoring

3. Draft replies + quotes

4. Auto-quote optional

5. Replies to strangers require approval

Done criteria

• Auth works for test account
• Post/reply/quote/like/retweet working
• Idempotency prevents dupes
• 30-min inbox loop runs without spammy behavior