The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to generate a shell script (scripts/conductor-setup.sh), apply executable permissions using chmod +x, and configure conductor.json to execute this script. This is standard behavior for automating repository initialization.- [DYNAMIC_EXECUTION]: The skill generates a zsh script at runtime based on the detected project structure and executes it to prepare the workspace environment. According to standard assessment, script generation from known templates for local setup is considered a lower-risk dynamic execution pattern.- [INDIRECT_PROMPT_INJECTION]: The skill's workflow involves scanning the repository's file system to identify frameworks, package managers, and environment files. This ingestion of untrusted repository data creates a surface where a malicious project structure could influence the agent's configuration actions.\n
Ingestion points: Repository file system inspection (searching for framework-specific files and environment variables) as described in SKILL.md.\n
Boundary markers: No explicit boundary markers or instructions to ignore embedded instructions in discovered files are present.\n
Capability inventory: The skill possesses the ability to write files (conductor.json, conductor-setup.sh), modify file permissions (chmod), and execute shell commands through the Conductor configuration.\n
Sanitization: The generated conductor-setup.sh script includes basic checks to ensure source .env files exist before attempting to symlink them.- [CREDENTIALS_UNSAFE]: The skill specifically targets sensitive file paths including .env and .env.local. It automates the discovery and symlinking of these files from a persistent root to isolated workspaces. While these files often contain secrets, the skill facilitates local environment management without hardcoding credentials or performing network exfiltration.

conductor-setup