The Agent Skills Directory

[COMMAND_EXECUTION]: Path Traversal vulnerability detected in the spec.py helper script.
Evidence: The create_spec function in spec.py uses the template argument (provided via the --add flag) to construct file paths without sanitization: dest_file = spec_dir / f"{template}.md". Similarly, the get_template_path function constructs a read path using the same variable.
Impact: This allows for directory traversal (e.g., using ../../) which could result in writing or overwriting files outside the intended specification directories, although restricted to the .md extension.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content from untrusted workspace files.
Ingestion points: The skill reads metadata and document existence from files like requirements.md and solution.md within the .start/specs/ directory via the spec.py --read command.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when the agent processes the output of the metadata script.
Capability inventory: The agent can execute file system commands (via spec.py) and transition through workflow phases based on the content it reads.
Sanitization: The script sanitizes the feature_name parameter using a regular expression but fails to sanitize the template parameter, creating an exploitable path for injected instructions to perform unauthorized file operations.

specification-management