The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and modify documentation files (markdown/MDX) under the docs/ directory. These files represent an ingestion point for external data that could contain malicious instructions. The skill possesses significant capabilities, including the Bash tool and file system modification tools (Write, Edit). There are no explicit boundary markers or instructions for the agent to sanitize or ignore embedded instructions within the documents it processes, creating a surface for indirect prompt injection.
Ingestion points: Documentation files located in the docs/ directory.
Boundary markers: None identified for distinguishing documentation content from agent instructions.
Capability inventory: Access to Bash, Write, Edit, Read, Grep, and Glob tools.
Sanitization: No filtering or sanitization of documentation content is specified.
[COMMAND_EXECUTION]: The skill instructions and subagent brief template recommend the execution of various development and build commands, including npm run build, pnpm update-api-docs, ruff format, ruff check --fix, and gh pr create. These are standard tools for managing a documentation and API repository and are used within their expected context.
[EXTERNAL_DOWNLOADS]: The skill documentation mentions that the pnpm update-api-docs command downloads an openapi.json specification from a remote GitHub repository (raw.githubusercontent.com). This is a standard operation for keeping API documentation in sync with production specifications.
[DATA_EXFILTRATION]: The instructions require the agent to verify API request/response shapes against eu.cloud.agenta.ai using a 'saved key'. While this involves network activity and the use of sensitive credentials, the target domain is the official platform of the skill's authoring organization (agenta-ai), representing legitimate vendor functionality rather than an exfiltration attempt.

write-docs