The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use a local CLI tool named steward to perform protocol validation, schema checks, and claim verification on repository-local governance artifacts.
[EXTERNAL_DOWNLOADS]: The skill includes an installation command using npx to fetch the stewardship skill package from the author's (arenukvern) repository, which is a standard procedure for skill management within this ecosystem.
[PROMPT_INJECTION]: The instructions establish a 'Stewardship Protocol' to steer the agent's behavior between different operational modes (e.g., 'tool-execution' versus 'steward-presence'). These are internal governance constraints that promote evidence-based reporting and do not attempt to bypass core safety guardrails.
[DATA_EXPOSURE]: The skill manages a 'self-model' for agent continuity but implements a strict redaction policy. It explicitly prohibits the persistence of raw chats, secrets, credentials, or private relational memory, ensuring that only synthesized, reviewable governance state is stored.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process repository data to maintain stewardship state, with integrated safeguards.
Ingestion points: Data is read from paths including .steward/**, docs/**, and AGENTS.md via the --mode-events and --self-model command flags.
Boundary markers: The framework utilizes structured 'Mode Boundary' blocks and 'Readiness Notes' to clearly delimit governance logic from ordinary tool output.
Capability inventory: The skill utilizes the steward CLI and has the capability to write synthesized governance state to the filesystem.
Sanitization: The protocol mandates a redaction policy that filters out sensitive information and unreviewed material before persistence.

steward-continuity-boundary-lifecycle