session-heartbeat
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions involve harvesting outputs from other agents and reading from a persistent checkpoint file, which represents an ingestion surface for potentially untrusted data.
- Ingestion points: Harvesting step (Step 2) and Resume protocol (Step 1) which reads the checkpoint file.
- Boundary markers: The skill does not explicitly define markers but references
orchestrating-agent-delegationfor isolation rules. - Capability inventory: Performs local file writes (checkpointing) and Git read operations (
git log,git cat-file). - Sanitization: Explicitly instructs the agent to "Verify [the checkpoint] against the tree; do not trust it."
- [EXTERNAL_DOWNLOADS]: The skill references documentation and repository context from Anthropic and the author's own GitHub repositories for versioning and protocol specifications.
- [COMMAND_EXECUTION]: Recommends the use of standard Git commands (
git log,git cat-file) to verify the repository state and ensure commit integrity across session resumes.
Audit Metadata