The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow defined in workflows/repo-analysis.md instructs the agent to execute shell commands like npm install -g repomix and repomix --output repomix-output.xml. Global package installation modifies system state and can be abused if the agent is directed to malicious packages.- [EXTERNAL_DOWNLOADS]: The skill fetches documentation from context7.com and clones repositories from external URLs using git clone. While GitHub is a recognized service, the skill allows cloning from arbitrary repositories provided by the user or extracted from documentation.- [REMOTE_CODE_EXECUTION]: The combination of downloading external code via git clone and processing it with tools like repomix constitutes a remote code execution risk, especially when the source is untrusted or determined at runtime.- [DATA_EXFILTRATION]: The skill loads sensitive environment variables such as GITHUB_TOKEN, GEMINI_API_KEY, and CONTEXT7_API_KEY through scripts/utils/env-loader.js. The presence of these credentials combined with the ability to make network requests via https.get and shell commands creates a path for potential secret exfiltration.- [PROMPT_INJECTION]: The skill processes untrusted external data from llms.txt files and source code repositories, creating a surface for indirect prompt injection.
Ingestion points: scripts/fetch-docs.js (fetching from context7.com) and workflows/repo-analysis.md (repository cloning).
Boundary markers: Absent; the skill does not instruct the agent to ignore instructions embedded in the documentation or source code it analyzes.
Capability inventory: Network requests (https), file system operations, and shell command execution (git, npm, node).
Sanitization: No sanitization is performed on the content of the documentation or code before it is provided to the agent context.

docs-seeker