The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation instructs the agent to execute npm install within the scripts/ directory, which downloads dependencies from the NPM registry, a well-known and reputable service.
[COMMAND_EXECUTION]: Several shell commands are used for automation within the skill's process: git fetch and git show are utilized to read specification files directly from the repository's origin/main branch; node scripts/render-guide.js is used to transform markdown into HTML; and grep is used to perform a stylistic check on the generated markdown output.
[INDIRECT_PROMPT_INJECTION]: The skill functions as a documentation generator that ingests untrusted user data, including rough funds flows and partner-specific notes. This creates a surface for indirect prompt injection where malicious instructions could potentially be embedded in the source data and reflected in the output guides. Ingestion points: Rough funds flows, numbered step lists, and partner implementation notes (SKILL.md). Boundary markers: No explicit delimiters or boundary markers are defined to isolate untrusted input data from the prompt's instructions. Capability inventory: Access to shell execution for git operations, Node.js script execution, and grep-based filtering (SKILL.md). Sanitization: No technical sanitization of input data is described; the skill only performs stylistic checks on the generated content.
[DYNAMIC_EXECUTION]: The skill uses a local Node.js script to process generated markdown content into HTML. This step constitutes dynamic execution as the script operates on files containing user-influenced data, and references to Playwright artifacts suggest the involvement of browser-based rendering components.

partner-dev-guide