The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes task evidence to create permanent lessons and heuristics via tools like loom_learn_record and loom_heuristic_propose. If this evidence is manipulated, it could inject persistent malicious instructions.\n- Ingestion points: Evidence and diagnostic findings recorded via the tools specified in SKILL.md.\n- Boundary markers: The skill does not employ explicit delimiters or boundary instructions to prevent the agent from obeying instructions embedded in the stored data.\n- Capability inventory: The skill relies on loom_learn_record, SynaBun_remember, loom_heuristic_propose, and loom_heuristic_review to retrieve and store information.\n- Sanitization: No specific data sanitization or filtering is defined for the ingestion process, relying instead on high-level reviewer confidence levels.

loom-learning