openclaw-message-stream

Plugin that scans OpenClaw session transcripts and emits structured findings for keyword, regex, PII, and sentiment signals.

The output is designed for AI agents, assistants, and runtime orchestrators that need clear message-level alerts.

Quick onboarding

Install

• Install from npm:

npm install --save-dev @yuenwinyun/openclaw-message-stream

• Or with pnpm:

pnpm add --save-dev @yuenwinyun/openclaw-message-stream

Add plugin to an OpenClaw host

Run in the OpenClaw host project:

npm install --save-dev @yuenwinyun/openclaw-message-stream

Then merge plugin settings into OpenClaw config:

node node_modules/@yuenwinyun/openclaw-message-stream/scripts/openclaw-config.mjs \
  --config /path/to/openclaw.config.json \
  --plugin-root /path/to/openclaw-host/node_modules/@yuenwinyun/openclaw-message-stream \
  --mode hybrid \
  --write

Reload OpenClaw and verify:

openclaw plugins list --json
openclaw plugins inspect openclaw-message-stream --json
openclaw plugins doctor

One-line starter config (optional)

From this package folder:

npm run openclaw:config

Runtime commands for agents

Use /msgstream in authorized sessions:

/msgstream
/msgstream one-shot
/msgstream one-shot --dry-run
/msgstream --mode scheduled --sessions session-a,session-b

Configuration

Set under plugins.entries.openclaw-message-stream.config.

• mode: hybrid, scheduled, streaming, one-shot (default hybrid)
• scan.limit: max sessions returned from session list
• scan.batchSize: max messages fetched per session scan
• scan.maxSessions: per-run session safety cap
• scan.intervalMs: poll interval for scheduled mode
• filters: label/agent/spawnedBy/search filters
• analysis: keyword + regex + PII + sentiment config
• gateway: OpenClaw gateway connection settings
• output: file/webhook/console output controls
• output.dryRun: analyze and checkpoint without external output

gateway.scopes defaults to include operator.read so the plugin can read sessions for scanning.

Emitted message record

Each emitted record contains:

• plugin
• mode
• session key
• message id and seq
• sender and role
• message content
• score and findings
• match boolean

Common outputs:

• console summary
• JSONL file output (optional)
• webhook POST payload (optional)

Notes

• dry-run mode still runs analysis and updates checkpoint state.
• Checkpoint deduplication is based on message id/sequence and content hash.
• Streaming mode needs valid gateway credentials/scopes.

Remove plugin safely

• Remove from OpenClaw config (plugin entry + allowed list + load paths).
• Uninstall package:

npm remove @yuenwinyun/openclaw-message-stream
# or
pnpm remove @yuenwinyun/openclaw-message-stream

AI skill support

This package also ships an OpenClaw/assistant skill at:

• skills/openclaw-message-stream/SKILL.md

You can copy it into your assistant skill folder for easier onboarding and common workflows.

Developer docs

For test, release, and publishing workflows, see:

• DEVELOPER.md