The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it is designed to ingest and analyze untrusted data from the user's workspace to generate content.
Ingestion points: The skill reads project source files (via sourcePaths), README files, and directory structures to understand the project and generate documentation.
Boundary markers: No specific delimiters or instructions to ignore embedded commands (e.g., "ignore all previous instructions") are defined when processing the project's source code or README content.
Capability inventory: The skill possesses significant capabilities, including executing shell commands via Bash, writing to the filesystem via Write, and invoking other skill tools like /doc-smith-images and /doc-smith-check.
Sanitization: There is no evidence of sanitization, escaping, or validation of the data ingested from the workspace before it is passed to the LLM for analysis.
[COMMAND_EXECUTION]: The skill performs various shell operations and script executions to manage its workspace and build documentation.
Evidence: The instructions involve executing git init and git commit within the .aigne/doc-smith/ directory. It also runs Node.js scripts (node skills/doc-smith-build/scripts/build.mjs) to generate navigation files and build HTML documentation. Additionally, it may perform npm install to ensure required dependencies like @google/genai are present.

doc-smith-create