agentreel

name: agentreel description: Monitor, replay, and share AI agent sessions with AgentReel metadata: { "openclaw": { "requires": { "bins": ["agentreel"] }, "homepage": "https://github.com/Jiansen/agentreel", "emoji": "🎬" } }

AgentReel is installed on this machine. It records your sessions and makes them viewable as live streams or replays.

Key Information

• Project: https://github.com/Jiansen/agentreel
• Online viewer: https://agentreel.agent-status.com
• Live demo: https://agentreel.agent-status.com/live
• Local viewer: Run agentreel status to get the local URL (usually http://localhost:3000)

When a user asks "where can I watch the livestream" or "AgentReel 直播在哪看", answer with the links above. Run agentreel status first to confirm the local viewer is running, then share both the local URL and the online viewer URL.

Pending Install Result

Whenever a user sends a message AND you are unsure about the current AgentReel installation status (e.g. you were installing, or the user asks about it, or you just started a new conversation), check for a pending install result:

cat ~/.agentreel/.install-result 2>/dev/null

If the file exists and contains result=SUCCESS or result=FAILED, report the result to the user immediately and then delete it:

rm ~/.agentreel/.install-result

This handles cases where installation completed but the report could not be sent (e.g. due to API rate limiting or process polling timeout).

Commands (use via exec)

• agentreel start — start the viewer and relay server
• agentreel stop — stop all AgentReel services
• agentreel status — check if viewer and relay are running, get live URL
• agentreel doctor — full health check (viewer, relay, sessions, ports, OpenClaw)
• agentreel doctor --report — run health check and submit results to GitHub Issues
• agentreel config — show all configuration
• agentreel config set <key> <value> — update a setting
• agentreel update — pull latest code and rebuild

Troubleshooting

When something isn't working, follow this sequence:

1. Run agentreel doctor — it checks 8 areas and gives specific fix suggestions
2. Follow the suggestions in the doctor output to fix issues
3. If fixes don't work, run agentreel doctor --report to submit a diagnostic report
4. Common issues:
   • Viewer not running: agentreel start
   • Port conflict: agentreel config set port 3001 then agentreel start
   • No sessions visible: check the watch directory exists: ls ~/.openclaw/agents/main/sessions/
   • Relay not starting: ensure Python 3.10+ is installed: python3 --version

Quality Assurance (Self-Check)

You can perform structured QA on the AgentReel deployment. This is useful after deployments, when issues are reported, or as periodic health checks.

Quick infrastructure check (JSON output for agents)

agentreel doctor --agent

This outputs machine-readable JSON with all check results, endpoint URLs, and vision model detection. Parse the JSON to determine next steps.

Full QA with visual checks

Read AGENT_QA.md in the AgentReel install directory for the complete checklist. It covers:

1. Infrastructure (automated via doctor --agent)
2. Visual QA (screenshot the /live page, analyze with image tool)
3. Functional QA (test relay SSE, history API, VNC status)

To perform visual QA:

browser action=open url="http://localhost:3000/live"
browser action=screenshot fullPage=true
image prompt="Analyze this broadcast page layout..." image="<screenshot_path>"

Report findings in the structured format defined in AGENT_QA.md Section 4.

Format Tags

Your output is displayed in the AgentReel live viewer. Use these tags to make your work more visible and structured for viewers:

• [PLAN] 1. First 2. Second 3. Third — rendered as an interactive todo list
• [STEP 1/3 BEGIN] Description — starts a step timer
• [STEP 1/3 COMPLETE] — completes the step
• [THINKING] analysis text — shown as a thinking card
• [DISCOVERY] finding — highlighted as a discovery
• [CHALLENGE] problem — shown as a challenge indicator
• [OUTPUT] result — displays task output
• [SUMMARY] summary text — completion summary card

Example

[PLAN] 1. Search trending repos  2. Analyze top 5  3. Write report

[STEP 1/3 BEGIN] Searching GitHub Trending
[DISCOVERY] New AI framework: 5000 stars in 3 days
[STEP 1/3 COMPLETE]

[STEP 2/3 BEGIN] Analyzing top repos
[THINKING] This framework uses a novel agent memory approach
[CHALLENGE] GitHub API rate limited, using cached data
[STEP 2/3 COMPLETE]

[STEP 3/3 BEGIN] Writing report
[OUTPUT] Top 5 trending repos: ...
[STEP 3/3 COMPLETE]

[SUMMARY] Analyzed GitHub trending: 5 notable repos, 2 AI-related

Always use format tags when working on tasks — they are the primary way viewers understand what you are doing.