The Agent Skills Directory

[DYNAMIC_EXECUTION]: The use of __import__ in scripts/arsenal.py is restricted to loading the local workspace module for aliasing purposes and does not process untrusted input or load modules from unknown paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied phase payloads and reviews which are later rendered into reports.
Ingestion points: JSON payloads read via the --input argument in command_apply and command_add_review in scripts/arsenal.py.
Boundary markers: The references/state-contract.md file explicitly states that generated Markdown and HTML are projections and should not be parsed back into state, reducing the risk of self-reinforcing injections.
Capability inventory: The skill is capable of writing files to the local system (atomic_write) but performs no network operations or subprocess execution of user input.
Sanitization: The scripts/render.py module uses html.escape to sanitize all user and agent text before rendering HTML, effectively mitigating XSS and instruction leakage in the rendered views.
[COMMAND_EXECUTION]: The skill does not execute arbitrary shell commands; it provides a structured CLI wrapper for its internal Python logic and does not use subprocess or os.system on untrusted data.
[DATA_EXFILTRATION]: No network calls or sensitive data access patterns were found. File operations are limited to the workspace JSON and report rendering within the specified project directory.

growth-arsenal-workspace