The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes local CLI tools such as codex, grok, and ffmpeg using subprocess.run and subprocess.Popen in several scripts including scripts/route_media.py, scripts/cli_media.py, and scripts/media_mp4.py. These executions are core to the skill's functionality as a router.
[DYNAMIC_EXECUTION]:
scripts/cli_media.py uses ctypes.CDLL(None) to access standard C library functions (renameat2, renamex_np) for cross-platform atomic file operations.
scripts/route_media.py contains a test seam that allows the execution of a script specified by the FORGE_ROUTE_MEDIA_FAKE environment variable.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted data from user-provided prompt files and reference images (e.g., in scripts/route_media.py).
Boundary markers: User prompts are wrapped in <art_direction> tags in scripts/cli_media.py before being passed to sub-agents.
Capability inventory: The skill performs network requests to provider APIs, writes files to the project directory, and executes subprocesses.
Sanitization: Logic in scripts/cli_media.py and scripts/generate_media.py scrubs secrets, URLs, and sensitive tokens from error messages and reports.
[EXTERNAL_DOWNLOADS]: scripts/generate_media.py performs network requests to download media from established providers including OpenAI, Google, xAI, BytePlus, and fal.ai.
[CREDENTIALS_UNSAFE]: The skill manages API keys for multiple providers, reading them from the environment or a configuration file in the user's home directory (~/.config/agent-sprite-forge/config.json). It isolates these credentials by stripping them from the environment when spawning sub-processes for local CLI routes.

generate2dmedia