Triage

Source of truth for Peter's moderation and email-triage policy.

Primary rules:

• TWITTER_MENTIONS.md
• TWITTER_REPLIES.md
• EMAILRULES.md

Also includes restore docs/templates for the current OpenClaw cron setup:

• shared clawblocker workspace templates: openclaw/clawblocker
• daily block digest templates: openclaw/blockdigest

Current Setup

There are 4 related automations:

1. X mention moderation
2. X reply-opportunity triage
3. Gmail triage
4. Daily X block digest email

The first 3 share one OpenClaw agent/workspace:

• agent id: clawblocker
• one workspace
• one repo checkout inside that workspace
• three runbooks
• separate state/audit files per cron
• twitter moderation also doubles as the birdclaw mention-cache warmer every 10 minutes; no second cache cron needed

The block digest stays separate because it is a simple daily mailer over the moderation audit log.

Shared Clawblocker Layout

Target layout:

~/.openclaw/
  workspace-clawblocker/
    AGENTS.md
    triage/                       # checkout of this repo
    jobs/
      twitter-moderation.md
      twitter-replies.md
      email-triage.md
    skills/
      bird-twitter-moderation/
        SKILL.md
    state/
      twitter.json
      twitter-replies.json
      email.json
    audit/
      twitter/
      twitter-replies/
      email/
  agents/
    clawblocker/
      agent/
        models.json
      sessions/
        sessions.json

Source-of-truth rule checkout:

• only one checkout: ~/.openclaw/workspace-clawblocker/triage
• runtime should read rules from that checkout
• runtime should not read ~/triage directly

Cron Jobs

1. Twitter Mention Moderation

• schedule: every 10 minutes
• agent: clawblocker
• model: openai/gpt-5.4
• thinking: high
• timeout: 360s
• delivery: none
• runbook: jobs/twitter-moderation.md
• also warms birdclaw mention cache before triage and records verified block/mute actions back into local SQLite

Message:

Run one clawblocker pass now using ~/.openclaw/workspace-clawblocker/jobs/twitter-moderation.md. Follow it exactly.

2. Twitter Reply Triage

• schedule: 17 */6 * * *
• timezone: Europe/London
• agent: clawblocker
• model: openai/gpt-5.4
• thinking: low
• timeout: 300s
• delivery: none
• runbook: jobs/twitter-replies.md
• warms birdclaw mention cache, runs a deeper xurl backfill, then surfaces only high-signal unreplied mentions worth Peter's time

Message:

Run one clawblocker pass now using ~/.openclaw/workspace-clawblocker/jobs/twitter-replies.md. Follow it exactly.

3. Email Triage

• schedule: 5 */3 * * *
• timezone: Europe/London
• agent: clawblocker
• model: openai/gpt-5.4
• thinking: low
• delivery: none
• runbook: jobs/email-triage.md

Message:

Run one clawblocker pass now using ~/.openclaw/workspace-clawblocker/jobs/email-triage.md. Follow it exactly.

4. Daily X Block Digest

• schedule: 0 9 * * *
• timezone: Europe/London
• agent: blockdigest
• model: openai/gpt-5.4
• thinking: low
• timeout: 180s
• delivery: none
• source audit: ~/.openclaw/workspace-clawblocker/audit/twitter/*.md

Required Local Tools

Twitter moderation

• birdclaw
• bird
• optional fallback: bird-gui
• active auth/cookies for X

Expected binary path in templates:

$HOME/Projects/birdclaw
$HOME/Projects/bird/bird

Expected runtime for birdclaw CLI today:

fnm exec --using 25.8.1 pnpm cli ...

Email triage / digest

• gog
• authenticated Gmail account

Expected default account:

steipete@gmail.com

Restore On A Fresh Machine

1. Clone this repo somewhere convenient, for example:

git clone https://github.com/steipete/triage.git ~/triage

2. Create the shared OpenClaw agent:

node ~/clawdbot/dist/index.js agents add clawblocker --workspace ~/.openclaw/workspace-clawblocker --non-interactive

3. Put this repo inside the workspace as the only runtime checkout:

git clone https://github.com/steipete/triage.git ~/.openclaw/workspace-clawblocker/triage

4. Copy template files from this repo into the live workspace:

• openclaw/clawblocker/AGENTS.md
• openclaw/clawblocker/jobs/twitter-moderation.md
• openclaw/clawblocker/jobs/twitter-replies.md
• openclaw/clawblocker/jobs/email-triage.md
• openclaw/clawblocker/skills/bird-twitter-moderation/SKILL.md
• openclaw/clawblocker/state/twitter.json.example
• openclaw/clawblocker/state/twitter-replies.json.example
• openclaw/clawblocker/state/email.json.example

5. Create the block digest workspace/agent from the templates in openclaw/blockdigest.

6. Add the cron jobs with the settings above.

7. Verify:

• twitter run reads jobs/twitter-moderation.md
• reply triage run reads jobs/twitter-replies.md
• email run reads jobs/email-triage.md
• both use agentId: clawblocker
• only one triage checkout exists under cron workspaces
• block digest reads ~/.openclaw/workspace-clawblocker/audit/twitter

Notes

• Rule edits belong in this repo first.
• After changing rules, pull the workspace checkout.
• State and audit logs are runtime data; do not store them in this repo.
• Email triage may auto-archive only exact Rule 1 / Rule 2 junk. Everything else remains observe-first.