AgentNotes for OpenClaw#

Send high-level job summaries to AgentNotes. Raw logs are rolled up hourly/daily into SparkNotes; you do not need to store every line forever.

Setup (once)#

1. AgentNotes dashboard → create agent (note the slug) → copy env vars from Connect.
2. Install this skill (see dashboard Connect → OpenClaw tab, or run install-skill.ps1 / install-skill.sh from the repo).
3. Add env to ~/.openclaw/openclaw.json:

json5

{
  skills: {
    entries: {
      agentnotes: {
        enabled: true,
        env: {
          AGENTNOTES_API_KEY: "an_...",
          AGENTNOTES_AGENT_ID: "your-slug",
          AGENTNOTES_BASE_URL: "https://your-app.vercel.app",
        },
      },
    },
  },
}

1. Restart OpenClaw / new session. Verify:

bash

node {baseDir}/scripts/verify.mjs

Recommended: one command per task#

Always pass a plain-English summary (this becomes the CEO-facing SparkNote):

bash

node {baseDir}/scripts/task.mjs --summary "Replied to 2 Telegram chats, scheduled 1 meeting"

With an extra step log:

bash

node {baseDir}/scripts/task.mjs --summary "Inbox sweep done" --message "Processed 12 threads" --step inbox

On failure:

bash

node {baseDir}/scripts/task.mjs --summary "Send failed" --failed --error "SMTP timeout"

Manual multi-step flow#

bash

export AGENTNOTES_RUN_ID=$(node {baseDir}/scripts/start-run.mjs)
node {baseDir}/scripts/log.mjs "Tool: calendar" --step tools
node {baseDir}/scripts/complete-run.mjs --summary "Updated 3 events"

When to log#

• After handling a user message on any channel
• After a cron / heartbeat completes
• After a skill or tool chain finishes (success or failure)

Dashboard#

$AGENTNOTES_BASE_URL/agents — SparkNotes tab is the long-term record.