The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it is designed to ingest and process documentation and code from untrusted external sources (web pages and GitHub repositories) without implementing safety boundary markers or instructions for agents to ignore instructions embedded in the data.
Ingestion points: Technical documentation fetched via WebFetch, content processed by Explorer and Researcher agents, and files extracted from cloned repositories using Repomix (e.g., repomix-output.xml).
Boundary markers: Absent. The instructions do not specify delimiters or warnings to prevent the agent from following instructions found within the processed documentation.
Capability inventory: The skill can execute shell commands (git clone, npm install, repomix), perform web searches, and fetch remote content.
Sanitization: No sanitization, filtering, or validation of the ingested external content is mentioned or implemented.
[COMMAND_EXECUTION]: The skill uses shell commands to manage its environment and process data.
Evidence: The skill attempts to install a global package using npm install -g repomix if the tool is not found.
Evidence: It performs repository cloning using git clone [repo-url] /tmp/docs-analysis to perform deep codebase analysis.
[EXTERNAL_DOWNLOADS]: The skill relies on fetching content from various third-party domains and repositories discovered at runtime.
Evidence: Regular use of WebFetch and WebSearch to retrieve content from documentation hosting platforms (e.g., Read the Docs, GitBook, Docusaurus).
Evidence: Clones external GitHub repositories for source code documentation extraction.

docs-seeker