A specialized skill for managing and monitoring the progress of other AI agents through automated check-ins and evidence-based reporting.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-supervision
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install agent-supervision using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Agent Supervision?
Agent Supervision is a lightweight, evidence-first framework designed to monitor the performance of Openclaw Skills in multi-agent environments. It provides a structured way to oversee secondary agents using fixed-interval check-ins or ETA-based follow-ups, ensuring tasks move toward completion without human intervention.
By focusing on compact status codes and delivery verification, this skill prevents stagnation and enforces accountability. It operates primarily through isolated bridge crons, allowing for high-frequency supervision that stays out of the primary user chat to maintain a clean, professional workspace.
Agent Supervision Use Cases
- Setting up recurring progress nudges every N minutes for long-running tasks.
- Enforcing ETA-based follow-ups where an agent must commit to a specific goal and deliverable.
- Detecting stagnation and triggering escalation protocols after repeated lack of progress.
- Managing cross-session messaging where synchronous delivery signals are unreliable.
- Implementing background supervision that remains hidden from the main user-facing conversation.
How Agent Supervision Works
- Context Acquisition: The supervisor reads the last 5-8 messages from the target session to identify the current goal and promised ETA.
- Evidence Evaluation: It analyzes the history for new deliverables, file changes, or updated version increments since the last check-in.
- Status Classification: The agent assigns a compact status code (A, P, W, R, or E4) based on whether the target is progressing or stagnant.
- Instructional Feedback: A short supervision message is sent to the target agent to prompt the next action or request a blocker list.
- Delivery Verification: In the event of a timeout, the supervisor re-inspects the target session to confirm the message landed successfully.
Agent Supervision Setup
To deploy Agent Supervision within Openclaw Skills, configure an isolated cron job with limited tool access. Use the following bash-like JSON configuration for the agent turn:
{
"kind": "agentTurn",
"message": "Read recent target messages, classify status, and send supervision line.",
"timeoutSeconds": 120,
"toolsAllow": ["sessions_send", "sessions_history"],
"thinking": "off",
"lightContext": true
}
Agent Supervision Data Schema & Taxonomy
The skill organizes supervision data using a compact taxonomy to minimize token usage and maximize clarity:
| Code | Meaning | Description |
|---|---|---|
| A | Acknowledged | Plan received, ETA has not yet expired. |
| P | Progress | Strong evidence of new deliverables or output. |
| W | Weak | Minimal changes (e.g., version increment only). |
| R1-R3 | Stagnant | Repeated checks with no new progress detected. |
| E4 | Escalation | Forced threshold; requires immediate evidence or blocker list. |
Agent Supervision Advanced Features
- Isolated Bridge Mode: Runs supervision in a separate thread to prevent noise in user-facing sessions.
- Sync-Timeout Recovery: Advanced logic to verify message delivery even when network timeouts occur.
- Hybrid Supervision: Support for fixed-interval cron jobs that dynamically adjust based on ETA-aware logic.
- Evidence-First Guardrails: Rejects narrative updates (vibes) in favor of verifiable technical progress.
- Custom Escalation Thresholds: Configurable triggers for when a stagnant agent requires higher-level intervention.
SKILL.md
Loading
Related Openclaw Skills
Logvance
A CLI utility that automatically converts UTC or ISO-formatted log timestamps to your local system time for easier debugging.
SEO (Site Audit + Content Writer + Competitor Analysis)
An all-in-one SEO specialist agent designed to handle technical audits, strategic content creation, and comprehensive ranking improvements.
Darcy Strategic Brain
A sophisticated strategic reasoning engine that utilizes multi-agent dialectics and statistical data analysis to validate complex business decisions.
NSFC Writing Guide
A professional AI collaboration tool designed to assist researchers in drafting, outlining, and refining National Natural Science Foundation of China (NSFC) grant applications.
QuickCSV
A high-speed utility for transforming JSON, YAML, and TSV files into standardized CSV formats with automatic header detection.
Perplexity Research ZH
A specialized research skill that utilizes Perplexity Sonar models via AISA to generate comprehensive, cited answers for deep academic or technical inquiries.
Featured*
