The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, which creates a surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through image analysis via the Read tool and web page capture via the /browse skill.
Boundary markers: The instructions include a specific 'Prompt Injection Boundary' section in SKILL.md that explicitly directs the agent to treat all text extracted from images and pages as descriptive data rather than instructions.
Capability inventory: The skill has the capability to write files, execute git commands, run local npm scripts, and use the GitHub CLI to create pull requests.
Sanitization: The skill employs schema-based validation via npm run validate:designs to ensure the integrity of the generated design JSON.
[COMMAND_EXECUTION]: The skill executes various shell commands to manage its workflow and interact with the developer environment.
Evidence: Uses git remote -v to detect if the current environment is the authoritative sleek-ui repository.
Evidence: Executes npm run validate:designs to perform integrity checks on generated artifacts.
Evidence: Uses git checkout, git commit, git push, and gh pr create to automate the catalog contribution process.
[EXTERNAL_DOWNLOADS]: The skill references and generates links to external resources as part of the design extraction process.
Evidence: Fetches typography configuration and generates stylesheet URLs from Google Fonts' official services.

design-extractor