The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute a local state-management script called state-commit. It pipes JSON-formatted transactions to this script to update the work item store. This is the intended mechanism for persistence within the framework and involves local execution pathing for the script and repository root.- [PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because the complete subcommand reads external validation manifests and evidence files from the filesystem. If these files are authored by an untrusted entity, they could contain instructions intended to influence the agent's logic.
Ingestion points: File paths provided to --validation-manifest and --evidence-ref in the complete subcommand.
Boundary markers: None specified in the instructions for reading these files.
Capability inventory: Read for file access, Bash for executing the state-commit script, and Glob for file discovery.
Sanitization: None described; the skill primarily checks for a 'passed' status within the manifest.

execute