The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process information from external sources such as code comments, API schemas, and SDK documentation to update a conceptual graph.
Ingestion points: The agent is directed to read and analyze code comments, external surface definitions (API/SDK/CLI), and existing graph nodes in SKILL.md.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore potentially malicious instructions embedded within the processed code comments.
Capability inventory: The agent utilizes the iskron_orient tool and has the capability to write to the graph through the writing and weaving protocols.
Sanitization: There is no mention of sanitizing or validating the text extracted from code before it is integrated into the graph or used to influence agent decisions.
[DYNAMIC_EXECUTION]: The protocol for checking code against the graph (step 3a) involves using an inventory script to map callable paths.
Evidence: The text states, "The inventory script lives in the consumer's repository and has the right to stand as a CI gate."
Risk: If the agent is expected to execute this repository-local script to perform the inventory, it constitutes a dynamic execution of potentially untrusted code found within the target project.

reconcile