The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: An automated scanner flagged a curl | sh pattern involving an external URL.
This string is located in scripts/test-dependency-install-guard.sh and is used exclusively as a test case to verify that the skill's security hooks correctly identify and block malicious remote code execution attempts.
[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.Popen and subprocess.run to manage worker agent CLIs.
These calls are implemented using argument arrays (shell=False) to prevent injection and are further protected by the validate-worker-command.py script, which enforces a strict identity gate for backend executables.
[DATA_EXPOSURE]: The skill includes proactive measures to prevent the leakage of sensitive data in logs.
The redact_sensitive_stream function in scripts/sentinel.sh and scripts/pm-orchestrate.sh uses regular expressions to detect and mask API keys (e.g., Anthropic, OpenAI, Gemini), bearer tokens, and authorization headers before they are written to the sentinel log files.
[OBFUSCATION]: Base64 encoding is utilized in scripts/dependency-install-guard.py and scripts/pm-orchestrate.sh.
This is used as a standard method for transporting structured JSON authorization snapshots through environment variables rather than as a technique to conceal malicious behavior.
[INDIRECT_PROMPT_INJECTION]: The skill orchestrates data between different agent sessions, which represents a potential attack surface.
This risk is mitigated by the skill's architecture, which enforces physical worktree isolation and utilizes PreToolUse hooks to intercept and block out-of-scope file edits at the runtime level.

multi-agent-orchestration