context-handoff
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read repository declarations and active work items (such as GitHub issues) to determine handoff policy and record state. These external sources could be manipulated by an attacker to include malicious instructions.
- Ingestion points: Repository declarations, GitHub issue bodies/comments, and local handoff files (SKILL.md).
- Capability inventory: Git operations (commit, push), GitHub CLI operations (posting comments), and feedback submission via
submit-feedback. - Boundary markers: The skill does not explicitly instruct the agent to use delimiters or ignore instructions found within the work items.
- Sanitization: No sanitization steps are defined for the content read from external work items before processing.
- [COMMAND_EXECUTION]: The instructions involve executing shell commands for version control (git) and interaction with GitHub (gh). The skill includes defensive instructions, such as verifying repository policies before pushing and avoiding
git add -Ato prevent unintentional staging of files. - [DATA_EXFILTRATION]: The skill involves sending task state, findings, and code checkpoints to external locations like GitHub repositories or issue trackers. While this is the intended purpose of the skill, it represents a data movement path. The skill mitigates this by requiring explicit repository or user authorization before performing remote writes.
Audit Metadata