The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The skill workflow requires executing multiple local Node.js scripts (detect-topic.js, fetch-docs.js, analyze-llms-txt.js) via the command line. While the scripts themselves are provided, this pattern increases the attack surface if the script logic is compromised.
EXTERNAL_DOWNLOADS (MEDIUM): The workflows/repo-analysis.md file instructs the agent to run git clone [repo-url] and npm install -g repomix. Executing these commands on untrusted repository URLs or installing global packages can lead to remote code execution if the sources are malicious.
DATA_EXFILTRATION (LOW): The scripts/utils/env-loader.js script recursively searches parent directories (up to .claude/.env) for .env files. This behavior can lead to the accidental exposure of API keys or secrets intended for other skills or agents if they are stored in shared parent directories.
INDIRECT_PROMPT_INJECTION (LOW): The skill is designed to fetch and process llms.txt files from external sources via context7.com.
Ingestion points: scripts/fetch-docs.js (fetches external documentation content).
Boundary markers: Absent; the content is piped directly into analysis scripts.
Capability inventory: Subprocess calls for git, npm, and node scripts.
Sanitization: Limited; analyze-llms-txt.js uses regex to extract URLs but does not sanitize the remaining content for instruction-like patterns.

docs-seeker