The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes a shell script scripts/fetch-stitch.sh that utilizes curl to download HTML design assets from a remote URL provided as a command-line argument. This is consistent with the skill's documented purpose of importing design files.
[COMMAND_EXECUTION]: The skill defines several command execution surfaces:
The scripts/fetch-stitch.sh script executes shell commands (mkdir, curl) to manage local directories and retrieve remote content.
The package.json file defines scripts for validation and fetching, which execute node and bash respectively.
These executions are scoped to the primary functionality of the skill (fetching and validating code).
[DATA_EXFILTRATION]: While the skill performs network requests via curl, the destination URL is an input parameter intended for fetching design specifications. The SKILL.md contains explicit anti-patterns forbidding the reading or storage of Stitch API keys or credentials.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and process untrusted external data (Stitch HTML designs):
Ingestion points: scripts/fetch-stitch.sh downloads HTML content into the workspace.
Boundary markers: None identified in the provided scripts for the input data.
Capability inventory: The skill has the capability to write files and execute local scripts (scripts/validate.js) using Node.js and @swc/core.
Sanitization: The scripts/validate.js script provides a level of sanitization by ensuring the generated output does not contain web-only HTML elements or hardcoded colors, enforcing the conversion to native primitives.
[DYNAMIC_EXECUTION]: The scripts/validate.js utility uses the @swc/core library to parse generated TypeScript/JSX code into an Abstract Syntax Tree (AST). This is used for static analysis and validation purposes rather than dynamic execution or eval() of the code.

stitch-react-native