deadline-prep
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes standard git commands (
git logandgit diff) to collect data about the current work session. These are read-only operations restricted to the local repository. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of git commit messages and log entries from
.claude/critical_log_changes.csv. - Ingestion points: Reads git history and a local CSV file.
- Boundary markers: None explicitly defined in the prompts, though the output is structured as a Markdown template.
- Capability inventory: No network operations, arbitrary code execution, or privilege escalation capabilities are present.
- Sanitization: No specific sanitization of log content is performed before interpolation into the demo outline. An attacker who can influence commit messages could potentially insert instructions aimed at the agent, though the impact is limited to the generated text file.
Audit Metadata