The Agent Skills Directory

[PROMPT_INJECTION]: The script scripts/add_retry_logic.py is vulnerable to indirect prompt injection.\n
Ingestion points: Command-line arguments --retriable and --permanent in scripts/add_retry_logic.py are used to define exception lists for retry blocks.\n
Boundary markers: Absent. There is no validation to ensure inputs are valid identifiers or safe strings, and no delimiters are used to separate user data from generated code.\n
Capability inventory: The script uses Path.write_text to modify source code files and adds imports from the vendor-owned project_watch_mcp package.\n
Sanitization: Absent. Exception lists are joined and injected directly into a Python code template via f-strings before being parsed by ast.parse. This allows a malicious input to break out of the intended syntax (e.g., by including a closing parenthesis and new lines) and inject arbitrary Python commands into the target file.\n- [REMOTE_CODE_EXECUTION]: The skill utilizes dynamic code generation and filesystem modification to automate codebase updates.\n
The script scripts/add_retry_logic.py assembles Python code blocks as strings and parses them into AST nodes at runtime.\n
The modified code is written back to local files using ast.unparse and Path.write_text. While this is the primary purpose of the skill, the generation process lacks security controls to prevent unintended code injection if the agent is directed by an untrusted source to modify a file.

implement-retry-logic