The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from external deliberation frames.
Ingestion points: The skill instructions command the agent to read 'typed hole + retrieved context' from frames listed via the acc frames command.
Boundary markers: There are no explicit instructions or delimiters provided to isolate the retrieved context from the agent's core instructions.
Capability inventory: The skill utilizes the acc_act tool to submit proposals and continue executions based on the processed data.
Sanitization: No sanitization, validation, or filtering of the retrieved context is mentioned or implemented in the instructions.

accint-frames