The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx ctx7@latest as a fallback mechanism to execute the Context7 CLI from the npm registry. This is governed by a strict execution policy that mandates explicit user approval for every request and encourages version pinning after the initial invocation to ensure session stability.
[COMMAND_EXECUTION]: The skill performs shell-based operations for environment probing (e.g., command -v ctx7) and documentation retrieval. These actions are protected by explicit user confirmation steps and cross-environment detection logic (supporting bash, PowerShell, and cmd.exe) to ensure commands are executed as intended.
[DATA_EXFILTRATION]: Queries sent to the Context7 service transmit library names and search terms over the network. The skill enforces mandatory pre-query redaction of sensitive data, including API keys, tokens, PII, and proprietary code, and requires disclosure to the user before transmission.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest third-party documentation, which presents a surface for indirect injection. It mitigates this risk with comprehensive instructions to treat all fetched content as data rather than instructions, mandatory visual delimiters (e.g., blockquotes) for external text, and a policy to discard suspicious or adversarial results.
[PERSISTENCE]: The skill includes functionality for managing agent skills, which involve writing Markdown files to local directories. This capability is restricted by mandatory user approval for every write operation, requiring confirmation of the target directory and the specific files being modified.

context7-expert