The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied text and external graph data to create or update persistent records, creating an attack surface for embedded instructions.\n
Ingestion points: User prompts (e.g., 'add node', 'record') and search results from tools such as iskron_search and iskron_semantic_search described in the 'Gate of Novelty' section of SKILL.md.\n
Boundary markers: The methodology defines a strict ontology of node types (given_as) and behavioral states (epistemic, ontic, and volitional modes) that constrain the interpretation of ingested data.\n
Capability inventory: The skill utilizes write-access tools including iskron_add_kriya, iskron_add_phenomenon, iskron_add_vimarsha, and the multi-operation iskron_batch.\n
Sanitization: Instructions prohibit the inclusion of provenance data in node bodies and restrict the attrs field to structured data (non-prose), reducing the surface for arbitrary instruction injection.\n- [COMMAND_EXECUTION]: The skill leverages the iskron_batch tool to execute sequences of graph modification commands. Although these are domain-specific operations for the Iskron graph, they provide a powerful mechanism for altering the persistent state and future context of the agent.

writing