ClawNetClawNet

Persona — Writing Style & Social Intelligence

Operations

0. Set Up X Token (when needed)

IMPORTANT: If any script fails with TOKEN_SETUP_NEEDED, you MUST run this setup flow before retrying. Do not just report the error — fix it.

Uses agent-browser for browser automation (same pattern as npm-publish skill).

Step 1: Open the developer portal

${SKILL_DIR}/scripts/setup-token.sh navigate --username <handle>

Status codes:

  • ALREADY_VALID — token works, nothing to do
  • NAVIGATED + PAGE_SNAPSHOT + APPS_FOUND — portal is open, proceed to Step 2
  • NOT_LOGGED_IN — tell user: "Log into X at x.com in Chrome, then I'll try again"
  • NO_BROWSER — agent-browser unavailable, fall back to manual instructions

Step 2: Navigate to Keys and tokens

The script outputs a page snapshot and list of apps. Use agent-browser to navigate:

AB="agent-browser --auto-connect"

# See the page structure
$AB snapshot -i

# Navigate to the app (use href from APPS_FOUND output)
$AB open "<app-url>"
sleep 3

# Find and click "Keys and tokens" tab
$AB snapshot -i
# Look for a link/tab containing "Keys and tokens" → get its ref
$AB click "@<ref>"
sleep 3

Adapt to what you see on the page. The portal structure is:

Dashboard → Projects list → Click app → "Keys and tokens" tab → Bearer Token

If the Bearer Token needs regenerating, find the Regenerate button and click it.

Step 3: Capture the token

Once on the Keys and tokens page:

${SKILL_DIR}/scripts/setup-token.sh capture --username <handle>

The capture script:

  1. Polls for up to 120 seconds looking for a Copy button
  2. Clicks Copy → reads token from clipboard (pbpaste)
  3. Validates the token against the X API
  4. Saves to ~/.claude/persona/tokens.json
  5. Clears clipboard for security

Status codes:

  • TOKEN_SAVED — success, retry the original operation
  • CAPTURE_TIMEOUT — couldn't find token, ask user to copy it manually

Manual fallback: If capture times out, ask the user to copy the Bearer Token value, then:

${SKILL_DIR}/scripts/save-token.sh --username <handle> --token <bearer_token>

1. Capture Writing Style

Build a writing profile from a user's X posts. Requires a valid X API token (stored in tokens.json or env var).

${SKILL_DIR}/scripts/capture.sh --username <handle> [--count 50] [--output <path>] [--refresh]
  • Fetches original English posts, scores by 60% recency + 40% engagement
  • Writes profile to .claude/persona/<username>.json
  • Cached for 7 days — pass --refresh to force re-fetch

2. Track Users

Manage a pool of X users to monitor. Pool stored in .claude/persona/pool.json.

# Add a user (validates via X API, auto-captures profile)
${SKILL_DIR}/scripts/track.sh add <username> [--note "reason"]

# Remove a user
${SKILL_DIR}/scripts/track.sh remove <username>

# List all tracked users with profile status
${SKILL_DIR}/scripts/track.sh list

# Refresh profiles for one or all users
${SKILL_DIR}/scripts/track.sh refresh [username]

3. Scan Social Intelligence

Run a social intelligence scan via xAI Grok. Requires XAI_API_KEY.

${SKILL_DIR}/scripts/scan.sh [--topics "Bitcoin SV, AI agents"] [--pool] [--save-topics] [--refresh]
  • Returns: Technical Developments, Content Opportunities, Notable Activity, Early Signals
  • --pool includes recent activity from tracked users
  • --save-topics persists topics to .claude/persona/topics.json
  • Cached for 4 hours — pass --refresh to force

4. Generate Draft (Full Pipeline)

Generate a styled draft post by combining all context layers — persona profile, body of work, git activity, social intelligence, and content strategy. Calls the Claude API directly.

${SKILL_DIR}/scripts/draft.sh [--profile <path>] [--scan <path>] [--topic "angle"] [--parts 3] [--output <path>] [--model claude-sonnet-4-6]

Context assembly (mirrors satchmo.dev pipeline):

  1. Voice examples from persona profile
  2. Body of work from .claude/persona/work.json
  3. Recent git commits from repos in work.json
  4. Social intelligence scan (cached or specify path)
  5. Content strategy rules

The LLM cross-references all layers — connecting trending topics to your actual work and recent commits. Requires ANTHROPIC_API_KEY.

5. Manage Body of Work

Configure the projects/products you've built — gives the LLM context about what to connect trending topics to.

# Add a project
${SKILL_DIR}/scripts/work.sh add --title "Project Name" --desc "What it does" --tags "tag1, tag2" [--repo owner/repo]

# List projects
${SKILL_DIR}/scripts/work.sh list

# Remove a project
${SKILL_DIR}/scripts/work.sh remove "Project Name"

Projects with a --repo field also feed into git activity fetching.

6. Fetch Git Activity

Fetch recent commits from configured repos (pulled from work.json repo fields + explicit repos).

${SKILL_DIR}/scripts/git-activity.sh [--repos "owner/repo1, owner/repo2"] [--per-repo 5] [--hours 48]

Uses GITHUB_TOKEN for private repos. Public repos work without auth.

7. Apply Style to Draft (Manual)

Assemble a style-matching prompt from a profile and draft content. Does NOT call an LLM — outputs a prompt payload for you to use. Use draft.sh instead for the full pipeline.

${SKILL_DIR}/scripts/apply.sh --draft <path-or--> [--profile <path>] [--format thread|single] [--max-chars 280]
  • Loads the persona profile, content strategy rules, and draft
  • Outputs JSON with system, prompt, and output_schema fields
  • Feed this to generateText() or your preferred model

8. Preview Post

Open a local preview of a styled post in the browser. Fully offline — no external services.

# Static preview (self-contained HTML, auto-opens browser)
${SKILL_DIR}/scripts/preview.sh --post <json-path> [--image <path>] [--username <handle>]

# Interactive playground (live editing, image generation, approval workflow)
bun run ${SKILL_DIR}/scripts/playground.ts --data <json-path> [--port 4747] [--open]
  • preview.sh: Static HTML preview with base64-embedded images
  • playground.ts: Live editor with real-time preview, image upload/generation, approve/reject workflow
    • --open flag opens browser automatically (default: no auto-open)
    • Shows real X avatar when token is available or profile has been captured
    • Generate Image button uses gemskills (requires claude plugin install gemskills@b-open-io)
    • Auto-exits 30s after browser tab closes (heartbeat-based)

Workflow: Generate a Post (Full Pipeline)

  1. Set up body of work: work.sh add --title "..." --desc "..." --tags "..." --repo "..."
  2. Capture voice profile: capture.sh --username <handle>
  3. Scan for intelligence: scan.sh --topics "..."
  4. Generate draft: draft.sh --parts 3 --output post.json
  5. Preview and edit: bun run playground.ts --data post.json --open
  6. Approve in playground

Or with a specific topic from the scan:

4b. draft.sh --topic "opportunity from scan" --parts 2 --output post.json

Workflow: Restyle an Existing Draft (Manual)

  1. Write a rough draft
  2. Run apply.sh --draft <path> --profile .claude/persona/<handle>.json
  3. Use the returned prompt payload with your preferred LLM
  4. Preview: preview.sh --post <output.json>

Workflow: Set Up Tracking Pool

  1. track.sh add wildsatchmo --note "self"
  2. track.sh add somedev --note "BSV builder"
  3. track.sh list — verify pool and profile status
  4. scan.sh --pool — include pool activity in scans

Storage

All data lives in .claude/persona/ in the project root:

.claude/persona/
├── <username>.json    # Individual writing profiles
├── tokens.json        # X API tokens keyed by username (multi-account)
├── pool.json          # Tracked user roster
├── work.json          # Body of work / projects config
├── topics.json        # Configured scan topics
└── last-scan.json     # Cached social intelligence scan

Token Resolution

X API tokens are resolved in this order:

  1. tokens.json by username — Multi-account: each persona has its own token in ~/.claude/persona/tokens.json (user-global, outside any repo)
  2. tokens.json any account — Falls back to any valid token from other accounts
  3. X_BEARER_TOKEN env var — App-only bearer token (legacy, single-account)
  4. X_ACCESS_TOKEN env var — OAuth 2.0 user token
  5. OAuth refresh — Using X_REFRESH_TOKEN + X_CLIENT_SECRET_ID
  6. TOKEN_SETUP_NEEDED — All methods exhausted, model should run setup-token.sh

Stored at ~/.claude/persona/tokens.json (user-global, never committed to any repo):

{
  "wildsatchmo": { "bearer": "AAA...", "added": "2026-03-16T19:00:00Z" },
  "bopen_io": { "bearer": "BBB...", "added": "2026-03-16T19:00:00Z" }
}

Environment Variables

Env vars are a fallback — prefer tokens.json for multi-account support.

Variable Required For Where to Get Notes
X_BEARER_TOKEN capture, track (fallback) https://developer.x.com/en/portal/dashboard Single-account fallback
X_ACCESS_TOKEN capture, track (fallback) OAuth 2.0 flow User token, works for reads too
X_REFRESH_TOKEN auto-refresh OAuth 2.0 flow Used with X_CLIENT_SECRET_ID
X_CLIENT_SECRET_ID auto-refresh https://developer.x.com/en/portal/dashboard OAuth client ID
XAI_API_KEY scan https://x.ai/api xAI Grok for social intelligence
ANTHROPIC_API_KEY draft https://console.anthropic.com/ Claude API for draft generation
GITHUB_TOKEN git-activity https://github.com/settings/tokens For private repo commit fetching