The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill transforms external HTML data from Stitch designs into React Native source code. This represents a vulnerability surface where malicious instructions embedded in design assets could influence the agent's code generation logic.
Ingestion points: HTML and image assets downloaded to the .stitch/ directory via scripts/fetch-stitch.sh as documented in references/workflow.md.
Boundary markers: Absent; the instructions do not define specific delimiters or "ignore" directives for the ingested HTML content.
Capability inventory: File system writes for component generation, execution of a local download script (bash), and AST parsing (node).
Sanitization: The skill contains explicit instructions in SKILL.md to redact PII (names, phones) from demo data and prohibits storing credentials or session cookies.
[COMMAND_EXECUTION]: The skill uses local scripts to automate its workflow. These include scripts/fetch-stitch.sh (a bash wrapper for curl) and scripts/validate.js (a Node.js script using @swc/core for AST validation). A Python test script (tests/test_fetch.py) also uses subprocess.run to verify the fetch utility.
[EXTERNAL_DOWNLOADS]: The skill fetches design assets from Google Cloud Storage URLs provided by the Stitch design tool. The fetch-stitch.sh script enforces HTTPS and uses hardened curl flags (e.g., --proto '=https') to mitigate download risks.

stitch-react-native