The Agent Skills Directory

[SAFE]: The skill implements a benign documentation workflow. It uses standard shell commands (ls, read_file, grep) to analyze codebase structure and content within user-uploaded directories. There is no evidence of unauthorized network access, credential harvesting, persistence mechanisms, or attempts to override agent safety protocols.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes untrusted source code and metadata to generate documentation summaries.
Ingestion points: Reads project files (e.g., package.json, pyproject.toml) and source code from the /mnt/user-data/uploads/ directory.
Boundary markers: Absent; the instructions do not explicitly provide delimiters or warnings to ignore malicious instructions embedded within code comments or string literals.
Capability inventory: Filesystem reads (ls, read_file, grep), file writing to /mnt/user-data/outputs/, and file presentation via present_files.
Sanitization: Absent; the skill does not specify any sanitization or validation of the code content before including it in the generated documentation.
Risk Assessment: The risk is assessed as safe because the skill's primary capabilities are limited to text generation and documentation formatting, which lack the high-impact capabilities (like arbitrary code execution or network exfiltration) typically targeted by indirect injection attacks.

code-documentation