The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run to coordinate its multi-stage pipeline, executing local Python scripts such as run_one_task.py and calling downstream skill scripts like aigc_textgen.py and aigc_imagegen.py. This behavior is observed in scripts/collection_pipeline/runner.py, scripts/collection_pipeline/single_task.py, and scripts/collection_pipeline/utils.py. These calls use sys.executable and target relative paths within the skill's own environment, which is a standard pattern for complex agent skill orchestration.
[DYNAMIC_EXECUTION]: The script scripts/collection_pipeline/single_task.py performs dynamic module loading using importlib.import_module. This is used to load shared parameter-building logic from the skill's own scripts/ directory at runtime, depending on the image type being processed. This mechanism is internal to the skill's logic and does not ingest code from external or untrusted sources.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as image URLs and user-provided descriptions, which are then used to build prompts for AI models. This creates a surface for indirect prompt injection.
Ingestion points: Untrusted inputs like imageUrls and imageDesc are processed in scripts/collection_pipeline/plan.py and scripts/collection_pipeline/single_task.py.
Boundary markers: The architecture uses specialized scripts (build_textgen_params.py, build_imagegen_prompt.py) to wrap user inputs in templates, effectively separating data from instructions.
Capability inventory: The skill possesses the capability to write files to session directories, execute internal scripts, and make network requests to vendor APIs.
Sanitization: The prompt construction scripts include specific logic to sanitize input and append sensitive word filtering instructions (e.g., SENSITIVITY_SUFFIX) to mitigate potential risks from the processed data.
[SAFE]: The skill includes an onboarding process (scripts/onboarding.py) for API key management. While it mentions modifying shell profile files (like .zshrc), these are instructions provided for the user to execute manually rather than automated actions by the agent. All network operations are directed at official vendor domains under linkfox.com.

linkfox-aigc-imagegen-cloth