The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and deliberate over brain_frame data retrieved from the acc_act tool. These frames are explicitly described in the instructions as 'untrusted task context,' which represents a vulnerability surface where malicious instructions embedded in external data could influence the agent's logic.
Ingestion points: Content returned from acc_act(runtime="solve") as described in SKILL.md.
Boundary markers: The skill lacks explicit technical boundary markers or instructions to ignore embedded commands within the ingested frames, although it does warn the agent that the context is untrusted.
Capability inventory: The skill utilizes the acc_act tool to submit proposals (runtime="continue") and close commitments (runtime="outcome"), providing a mechanism for injected instructions to trigger tool actions.
Sanitization: There are no instructions for sanitizing, escaping, or validating the text within the brain_frame before it is used to generate proposal_text.

accint-solve