The Agent Skills Directory

PROMPT_INJECTION (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8).
Ingestion points: Task descriptions and feature metadata loaded from a queue via execute_queue or execute_feature.
Boundary markers: None found. There are no explicit delimiters or instructions to ignore embedded commands within the feature tasks.
Capability inventory: The lifecycle includes IMPLEMENT (writing code), VALIDATE (running tests/validation), and COMMIT (Git operations), which provide a significant capability surface for an attacker providing a malicious task.
Sanitization: No evidence of input validation or content sanitization for the task data is mentioned in the documentation.
COMMAND_EXECUTION (LOW): The skill performs dynamic code generation and execution (Category 10).
The documentation explicitly describes a workflow where the agent writes code to pass tests and then executes those tests. This implies runtime execution of generated or modified scripts.
Severity is downgraded to LOW as this is the primary stated purpose of an 'autonomous coding' skill, although the underlying script scripts/task_executor.py was not provided to verify the presence of sandboxing or safety controls.

ac-task-executor