The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is explicitly designed to ingest untrusted data from external sources which could contain malicious instructions.
Ingestion points: External word sources including GitHub issues, README files, source code fragments, RFCs, and conversation transcripts (defined in SKILL.md).
Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions to prevent the agent from obeying instructions embedded within the ingested 'shabda' (content).
Capability inventory: The skill uses a suite of tools including iskron_look, iskron_orient, iskron_update, and iskron_semantic_search to modify the internal knowledge graph, and relies on a GitHub adapter (using gh) for data retrieval.
Sanitization: There is no mention of sanitization, escaping, or filtering of the external content before it is processed by the agent and written to the graph.
[COMMAND_EXECUTION]: The skill's architecture relies on external adapters to execute shell commands (e.g., using gh for GitHub access) and internal graph tools (iskron_*) to perform its primary functions. While these are presented as vendor-specific tools for the iskron-ai ecosystem, they represent the execution of local binaries based on external data inputs.

intake