The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The core workflow relies on the execution of three Node.js scripts (detect-topic.js, fetch-docs.js, and analyze-llms-txt.js) using the node command. These scripts are not included in the provided file, meaning their logic is opaque and unverifiable.\n- [CREDENTIALS_UNSAFE] (HIGH): The skill's environment documentation specifies that scripts load .env files from sensitive locations including .claude/skills/.env and .claude/.env. This grants the scripts access to potentially broad-scoped API keys and secrets not intended for this specific skill.\n- [EXTERNAL_DOWNLOADS] (MEDIUM): The fetch-docs.js script is designed to retrieve documentation from context7.com, which is not a trusted source. This introduces untrusted external data into the agent's execution environment.\n- [PROMPT_INJECTION] (HIGH): The skill is highly vulnerable to Indirect Prompt Injection. It processes llms.txt files from external web sources and uses analyze-llms-txt.js to recommend an "agent distribution strategy." An attacker-controlled llms.txt could contain instructions that manipulate the agent's logic or resource usage (e.g., forcing a "7 agents" strategy to consume tokens or execute malicious instructions).\n- [DATA_EXFILTRATION] (MEDIUM): Although no explicit exfiltration command is shown, the scripts have the necessary capabilities (network access via fetching and file access via .env loading) to exfiltrate sensitive data.

docs-seeker