The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses instruction overriding techniques by commanding the agent to operate in 'AUTONOMOUS MODE' and explicitly stating 'Do NOT ask questions'. This attempts to bypass the agent's default behavior of seeking user confirmation before performing significant actions like modifying project files or installing packages.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of various documentation-related packages (e.g., @fastify/swagger, swagger-ui-express, @nestjs/swagger) based on the detected framework. While these are well-known tools from established organizations, the dynamic nature of the installation based on code detection is a notable behavior.
[COMMAND_EXECUTION]: The skill requires the execution of package managers (npm, pip, etc.) to install dependencies and involves recursive file system scanning to extract API routes and schemas from the project source code.
[DATA_EXFILTRATION]: The skill writes execution metadata and telemetry to ~/.claude/projects/skill-telemetry.md. This involves writing data outside the immediate project directory to a hidden path in the user's home directory.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection.
Ingestion points: The skill parses the entire project API layer, including source code, configuration files, and comments across multiple languages (JS/TS, Python, Go, Ruby, etc.).
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded agent-targeted directives within the codebase being scanned.
Capability inventory: The skill possesses file-write capabilities (docs generation and telemetry) and the ability to trigger package installations via shell commands.
Sanitization: No sanitization or validation of the ingested code content is specified before it is processed or used to determine which packages to install.

api-docs