deadline-prep

Pass

Audited by Gen Agent Trust Hub on Mar 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses local shell commands to retrieve repository metadata. Specifically, it executes git log --oneline and git diff --stat to gather work history for summary. These commands are restricted to read-only operations on the project's local git history.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its data processing workflow.\n
  • Ingestion points: Untrusted data is ingested from .claude/critical_log_changes.csv and the output of the git log command (commit messages).\n
  • Boundary markers: The instructions lack explicit delimiters or safety warnings to prevent the agent from obeying instructions embedded in the logs or commit messages.\n
  • Capability inventory: The skill can read local files, execute git commands, and write the final report to .claude/demo-outline.md.\n
  • Sanitization: There is no evidence of sanitization, filtering, or escaping applied to the external log data before it is interpolated into the demo generation prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 17, 2026, 08:48 AM