The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references https://stitch-design.ai/docs/sdk/ai-sdk and https://stitch-design.ai/docs/design-md/overview in SKILL.md. These URLs are flagged as malicious/blacklisted by automated reputation scanners.
[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to install external software packages (using npm install or pip install) and explicitly state that the documentation on the blacklisted stitch-design.ai domain should be treated as the 'ground truth' for installation and SDK usage.
[COMMAND_EXECUTION]: The skill executes several shell commands, including open, xdg-open, and start to open local files in the browser. It also executes a zip command to create a delivery bundle. The project name, which is supplied by the user during the interview phase, is used directly in the shell command for zipping, creating a command injection surface.
[CREDENTIALS_UNSAFE]: The skill guides the agent to handle API keys stored in environment variables or .env files. While it includes a directive not to echo the keys, the keys are intended to be used with an SDK and infrastructure whose primary documentation source is a blacklisted domain.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its ingestion of untrusted data.
Ingestion points: User-provided product descriptions, brand adjectives, and image descriptions are collected during the Phase 1 interview in SKILL.md.
Boundary markers: No delimiters or 'ignore embedded instructions' warnings are present when processing these inputs.
Capability inventory: The skill has capabilities for shell command execution (zip, open) and network operations via the Stitch SDK.
Sanitization: There is no evidence of sanitization or validation of user-provided strings before they are incorporated into project titles, file paths, or generation prompts.

brand-landingpage