The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently utilizes subprocess.run to coordinate its multi-stage pipeline, calling local helper scripts and downstream generation tools. Evidence is present in scripts/collection_pipeline/utils.py, scripts/collection_pipeline/single_task.py, and scripts/collection_pipeline/runner.py.
[REMOTE_CODE_EXECUTION]: The module scripts/collection_pipeline/single_task.py uses importlib.import_module to load internal build scripts at runtime from computed filesystem paths, which represents dynamic code execution. Additionally, the skill's shared utilities in scripts/collection_pipeline/utils.py execute external scripts using the sys.executable environment.
[EXTERNAL_DOWNLOADS]: The skill's SKILL.md file directs the agent to download and extract an onboarding package from https://agent-files.linkfox.com/skills/linkfox-onboarding/release.zip if local maintenance documentation is missing. This domain is managed by the vendor (linkfox-ai).
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from user inputs and interpolates them into generative templates.
Ingestion points: User-provided text fields such as point, pointHint, and imageDesc are processed in scripts/collection_pipeline/plan.py and passed to downstream models.
Boundary markers: The prompt templates in scripts/collection_pipeline/templates/ utilize Markdown headers (e.g., ## 上下文) to separate data but do not implement hardened isolation mechanisms to prevent instruction override.
Capability inventory: The skill possesses capabilities to write files to the session directory (save_image_plan.py) and execute shell commands (subprocess.run).
Sanitization: There is no evidence of input sanitization, filtering, or escaping of user-supplied strings before they are interpolated into the final system prompts.

linkfox-aigc-imagegen-cloth