The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the Codex CLI using the subprocess module. It correctly uses list-based arguments for subprocess.run, which prevents shell injection vulnerabilities by avoiding the use of a shell interpreter for command execution.
[INDIRECT_PROMPT_INJECTION]: The skill parses project-level documentation (CLAUDE.md) and skill metadata to generate AGENTS.md. Ingestion points: Files are read from the project root and .claude/skills/ directory via claude_parser.py. Boundary markers: The skill does not implement specific boundary markers for the generated text, as its primary purpose is documentation generation for human and agent reference. Capability inventory: The skill has the ability to write files to the project root (AGENTS.md, CLAUDE.md) and execute codex commands via codex_executor.py. Sanitization: External configuration data is parsed using yaml.safe_load(), which prevents unsafe object instantiation during parsing.
[DATA_EXPOSURE]: The skill manages project documentation and does not attempt to access sensitive system files (e.g., SSH keys, environment secrets) or exfiltrate data to external domains.

codex-cli-bridge