The Agent Skills Directory

[DATA_EXFILTRATION]: The doc_generator.py script performs file read operations using paths derived from an external configuration file (marketplace.json). The script uses lstrip('./') on component paths, but this does not prevent path traversal attacks using sequences like subdir/../../ within the path. This could allow reading files outside the intended plugin directories if the marketplace metadata is compromised.
[COMMAND_EXECUTION]: The SKILL.md file provides integration examples that instruct the agent to use subprocess.run to execute the Python documentation generator script. This encourages the agent to use shell-level execution for task automation.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from marketplace.json and markdown files across the plugins/ directory to generate documentation.
Ingestion points: Data is read from .claude-plugin/marketplace.json and the frontmatter of various markdown files.
Boundary markers: None are employed to distinguish between data and instructions.
Capability inventory: The script has file-write capabilities, specifically creating files in the docs/ directory.
Sanitization: No escaping or sanitization is performed on the ingested content before it is rendered into documentation templates, which could lead to the generation of malicious documentation content if the source data is compromised.
[DYNAMIC_EXECUTION]: The doc_generator.py script includes a custom template engine (SimpleTemplate) that uses regular expressions and getattr() to dynamically resolve and inject values into templates at runtime. While limited in scope, this dynamic resolution pattern increases the attack surface for potential property injection or unexpected code behavior.

documentation-update