---

name: agent-canary description: Plant decoy credentials in your OpenClaw workspace to detect malicious skill behavior. Canary tokens trigger alerts when read, copied, or exfiltrated. Activate when user says "deploy canary", "set up honeypot", "plant decoy", "canary check", "canary status", or "canary cleanup". version: 1.0.0

Agent Canary

Decoy credentials planted in your workspace. When a skill touches them, you know.

When to Activate

• User says: "deploy canary", "set up honeypot", "plant decoy credentials"
• User says: "canary check", "canary status", "any triggers?"
• User says: "canary cleanup", "remove canary"
• User says: "canary report", "show incidents"

Do NOT activate for normal conversations about security.

Deploy Flow

When user asks to deploy canary:

1. Run python3 skills/agent-canary/scripts/plant_canaries.py
2. Report what was planted and where
3. Set up a cron job for monitoring (every 30 minutes):

cron add:
  name: agent-canary-monitor
  schedule: every 30 minutes
  payload: agentTurn
  message: "Run canary check. Execute: python3 skills/agent-canary/scripts/check_canaries.py. If any triggers found, immediately alert the user with details."
  delivery: announce
  sessionTarget: isolated

4. Tell user monitoring is active

Check Flow

When user asks for canary status or check:

1. Run python3 skills/agent-canary/scripts/check_canaries.py
2. Report results:
   • If clean: "All canaries intact. No triggers."
   • If triggered: Format alert with severity, type, file, and recommendation

Cleanup Flow

When user asks to remove canary:

1. Run python3 skills/agent-canary/scripts/cleanup_canaries.py
2. Remove the monitoring cron job (cron list, find agent-canary-monitor, remove it)
3. Confirm cleanup complete

Alert Format

When a trigger is detected, notify the user:

CANARY TRIGGERED

Severity: [CRITICAL/HIGH/MEDIUM] Type: [file_modified/file_deleted/token_in_log/file_accessed] File: [path] Detail: [description]

Recommendation: Check which skill was running at [timestamp]. Consider rotating real credentials as precaution.

Report Flow

When user asks for canary report:

1. Read ~/.openclaw/agent-canary/incidents.log
2. Summarize all incidents chronologically
3. Show current canary status (files planted, last check time)

Technical Details

Token Types

Type	Marker	Example
AWS Key	AKIACANARY	AKIACANARY7F3A2B9X0
GitHub PAT	ghp_CANARY_	ghp_CANARY_a1b2c3...
Stripe	sk_live_CANARY_	sk_live_CANARY_x9y8...
API Key	sk-CANARY-	sk-CANARY-e1f2a3...
DB Password	CANARY_PASS_	CANARY_PASS_Kx9m2Nv7

All tokens contain CANARY in the value, making them trivially identifiable as decoys.

Files Created on Deploy

File	Location	Purpose
.env.canary	Workspace root	Fake env config with canary credentials
secrets.backup.json	Workspace root	Fake credential backup
canary-tokens.json	memory/	Fake service credential cache
manifest.json	~/.openclaw/agent-canary/	Token registry and file hashes
incidents.log	~/.openclaw/agent-canary/	Incident history

Detection Methods

1. File hash comparison — detects modification of canary files
2. File existence check — detects deletion (possible cleanup after exfil)
3. atime monitoring — detects unexpected file reads
4. Log grepping — scans exec logs for canary token strings
5. Git diff — checks if canary files were staged by a skill