checkpoint-work
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local management utility located at .agents/universal-agent-skills/runtime/cli.mjs to manage session activation and commit checkpoints.
- [PERSISTENCE]: The skill is designed to maintain semantic state across different agent sessions by writing to the .agents/state/continuity/ directory.
- [INDIRECT_PROMPT_INJECTION]: The skill reads previously saved objectives and plans from markdown files, creating an ingestion surface for potentially influenced data. Ingestion points: .agents/state/continuity/work-items/ work-item-id /semantic.md. Boundary markers: The schema uses standard Markdown headers but lacks specific delimiters to isolate untrusted content. Capability inventory: File system writing and local CLI execution for state updates. Sanitization: The instructions specifically mandate the redaction of credentials, tokens, and personal data from all saved capsules.
Audit Metadata