The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes data from external websites, which is inherently untrusted. However, the skill provides robust guidance for managing this risk.
Ingestion points: Web extraction results from nimble extract, nimble search, and nimble map commands as described in SKILL.md and references/nimble-extract/reference.md.
Boundary markers: The skill includes a dedicated rules/output.md file titled "Handling Extracted Web Content" which explicitly instructs the agent to treat fetched content as "untrusted third-party data" and provides an "ignore embedded instructions" behavioral model.
Capability inventory: The skill utilizes Bash (with access to python3, uv, npm, mkdir, and standard file utilities) to process extracted data.
Sanitization: Instructions in rules/output.md mandate incremental reading (using head and grep) and shell redirection to files (> .nimble/file.md) rather than direct interpolation of large scraped contents into the prompt.
[DYNAMIC_EXECUTION]: The skill generates and executes Python and Playwright scripts to orchestrate parallel tasks and conduct site investigations.
Evidence: references/batch-patterns.md contains Python script templates that are written to local storage and executed with uv run to handle batch processing via the nimble CLI. references/nimble-extract/browser-investigation.md provides Playwright scripts for discovering site selectors.
Context: These dynamic execution patterns are used for the primary purpose of the skill (scraping orchestration and site investigation) and use secure subprocess execution methods.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of official vendor tools.
Evidence: rules/setup.md provides instructions for installing the Nimble CLI via npm i -g @nimble-way/nimble-cli and adding the Nimble MCP server from https://docs.nimbleway.com/mcp.
Context: All external resources are official tools provided by the skill author (Nimbleway).

nimble-web-expert