The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run and subprocess.Popen to execute shell commands, primarily for interacting with git repositories and AI agent client binaries (specifically 'claude' and 'codex'). These commands are used to verify repository cleanliness, check commit history, and query runtime status. While execution depends on paths defined in configuration or environment variables, the skill implements checks to ensure binaries are located in well-known or explicitly defined paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes project-specific data from user-controlled files like CONTEXT.md and REFERENCE.md. It extracts text from bolded metadata headers and checklists (e.g., Phase, Status, Next Actions) to populate context anchors for the AI agent. This ingestion point lacks explicit boundary markers or sanitization for the extracted content, which could allow instructions embedded in these files to influence agent behavior during project resumption.
[DYNAMIC_EXECUTION]: Test scripts within the skill, such as scripts/test_conformance.py, utilize importlib.util and exec_module to dynamically load and run the functional scripts under test. Additionally, scripts/conformance.py uses ast.literal_eval for configuration parsing. While appropriate for testing and configuration management, these patterns involve runtime execution of dynamically identified code components.
[DATA_EXPOSURE]: The skill reads and audits sensitive information including agent transcripts (.jsonl files) and coordination state. However, it incorporates a robust sanitization mechanism in scripts/capability_evidence.py using comprehensive regular expressions to identify and prevent the recording of authentication materials (such as API keys, bearer tokens, and private keys) into audit logs, demonstrating a security-aware design for data handling.

synthesis-agent-conformance