The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The installation script scripts/install.sh automatically attempts to discover and extract an Anthropic API key from the local file ~/.openclaw/agents/main/agent/auth-profiles.json. This key is then stored in a systemd service environment variable.
[DATA_EXFILTRATION]: The skill reads from a sensitive file path ~/.openclaw/agents/main/agent/auth-profiles.json which contains agent authentication data.
[COMMAND_EXECUTION]: The installer uses sudo to create a new systemd service file at /etc/systemd/system/iblai-router.service and executes systemctl commands to enable and start it, granting the skill persistence and elevated execution context.
[REMOTE_CODE_EXECUTION]: The README.md file suggests an insecure command pattern for users: curl -s http://127.0.0.1:8402/stats | python3. Piping output from a network request directly into a Python interpreter is a dangerous practice that can be exploited if the endpoint is compromised.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it analyzes the content of user messages to determine routing logic in server.js.
Ingestion points: server.js ingests message content from the request body forwarded by OpenClaw.
Boundary markers: No delimiters or safety instructions are used to separate user data from the classification logic.
Capability inventory: The skill controls which LLM model processes a request and proxies sensitive API payloads.
Sanitization: No sanitization or filtering is performed on the text used for keyword-based scoring.

iblai-router