The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a bundled Python script scripts/deploy-hermes.py to execute system commands such as ssh, scp, and docker compose to deploy and manage services on a target virtual machine.\n- [CREDENTIALS_UNSAFE]: The skill references sensitive file paths, specifically the SSH private key at ~/.ssh/cloud-weaver. Accessing this key is necessary for the skill's primary function of remote deployment via SSH.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided configuration (environment names and ports) which are then used in shell commands. The risk is mitigated by the following evidence chain:\n
Ingestion points: Parameters collected from user prompts in SKILL.md.\n
Boundary markers: None.\n
Capability inventory: Execution of ssh, scp, and docker via subprocess.run in scripts/deploy-hermes.py.\n
Sanitization: Strict validation using regex ([a-z0-9_]+) for environment names, IP address parsing for hostnames, and integer type enforcement for ports in scripts/deploy-hermes.py.\n- [SAFE]: The deployment implementation includes several security-positive patterns:\n
Secret Generation: Instead of using hardcoded credentials, the script generates high-entropy secrets (API keys, database passwords) locally using the secrets module.\n
Permission Hardening: The script ensures that sensitive configuration files (e.g., .env) on the target VM are created with restricted permissions (chmod 600).\n
Secure Command Invocation: The Python code uses list-based arguments with subprocess.run, which avoids the vulnerabilities associated with shell interpolation.

cloud-weaver-waha