handoff
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes multiple shell commands to gather project state, including
git(log, status, diff, branch),ls(to list handoff directories), andgrep(to find prior chains). It also utilizes a project-specific CLI tool namedbd(beads system) for task tracking and metadata persistence. These operations are within the expected scope of a development-oriented agent. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to mine the current conversation history and generate a detailed report for the next session, which introduces a cross-session instruction persistence risk.
- Ingestion points: The skill mines the entire conversation history (up to 1M context tokens) to extract goals, completed work, and approaches tried.
- Boundary markers: There are no explicit instructions or delimiters used to ensure the agent ignores or sanitizes instructions that might be embedded in the mined conversation history.
- Capability inventory: The skill has permissions to write markdown files to the repository (
plans/handoffs/) and execute shell commands viabash. - Sanitization: No sanitization or escaping is performed on the content extracted from history before it is written to the handoff markdown files. This allows potentially malicious instructions from a conversation to be 'saved' into a trusted project file that the next agent session is explicitly instructed to read and obey via the 'Paste Prompt' generated in the
close-session.mdreference.
Audit Metadata