The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to harvest sensitive information, including Gateway tokens, Feishu app IDs, and app secrets, by reading and searching the ~/.openclaw/openclaw.json configuration file.
[COMMAND_EXECUTION]: The skill provides procedures for the agent to execute shell commands like cat, grep, find, and ls to collect metadata, configuration secrets, and agent identities from the file system.
[REMOTE_CODE_EXECUTION]: The skill enables executing commands on remote instances through SSH, facilitating cross-instance management and data collection.
[DATA_EXFILTRATION]: The skill describes methods to transmit collected data and credentials between instances using sessions_send or by directly calling external APIs (e.g., Feishu) with harvested secrets.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect injection as it processes data from remote agents and SSH outputs without defined boundaries or sanitization.
Ingestion points: Messages received via sessions_send and outputs from remote SSH sessions.
Boundary markers: None present to delimit untrusted remote data.
Capability inventory: SSH command execution, file system access (cat, find), and network operations (curl, requests).
Sanitization: No evidence of input validation or escaping for remote content.

cross-team-comm