The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator script run-generate-story.py uses subprocess.run to execute helper scripts and system tools like ffmpeg and curl. Analysis shows these calls are implemented safely using argument lists rather than shell strings, which prevents shell-based command injection. The skill also uses tempfile.mktemp for temporary file paths, which is a deprecated practice but does not pose a direct security threat in this context.- [DATA_EXFILTRATION]: The skill communicates with api.replicate.com and bedtimestories.bruce-hart.workers.dev. Both are legitimate destinations: Replicate is a well-known AI service provider, and the workers.dev domain belongs to the skill's author. API tokens are correctly handled via environment variables and passed securely in request headers.- [PROMPT_INJECTION]: The skill takes user-provided content (story title and body) and interpolates it into prompts sent to Replicate for image and video generation. While this represents an attack surface for indirect prompt injection, the risk is minimal as it targets downstream media generation models.
Ingestion points: User-provided story prompt or source material processed in SKILL.md and passed to run-generate-story.py.
Boundary markers: None identified. Content is interpolated directly into prompt templates in run-generate-story.py.
Capability inventory: subprocess.run executes curl for API interactions and ffmpeg for video processing in run-generate-story.py.
Sanitization: The compact_story_excerpt function in run-generate-story.py limits the length of user text used in prompts, but no specific escaping or filtering for injection patterns is performed.

generate-story