The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a workflow for processing external tasks (vimarshas) received in an inbox. This represents a surface for indirect prompt injection. \n
Ingestion points: Tasks are ingested from the posed_to field in the agent's inbox, which can be populated by other agents or users (SKILL.md). \n
Boundary markers: The skill incorporates a 'Reality-Audit' protocol and 'Inquiry' responses to verify the state of the repository and the graph against the instructions, although explicit delimiters for untrusted text are not defined. \n
Capability inventory: The agent uses tools like iskron_orient and iskron_search to interact with a state graph and performs standard Git operations (commit, merge) within its defined scope. \n
Sanitization: Validation is performed post-execution through the 'Reality-Audit' process to ensure the implementation matches the intended state.\n- [COMMAND_EXECUTION]: The instructions direct the agent to perform standard Git repository operations, such as managing branches and committing code according to the project's 'repo-ritual' (SKILL.md, Section 3).

on-duty