Gemini Audit Workflow

When to use

Use this skill when you need an independent review of a task and its implementation plan before proceeding with execution. The auditor will validate alignment, flag concerns, and provide an APPROVE/REJECT decision.

Procedure

1. •

   Ensure .agent/task.md contains the current task definition:

   markdown

   # Current Task
   [Objective description]
   
   ## Acceptance Criteria
   - [ ] Criteria 1
   - [ ] Criteria 2
   

2. •

   Ensure .agent/implementation_plan.md contains the proposed plan:

   markdown

   # Implementation Plan
   [Plan overview]
   
   ## Steps
   1. [ ] Step 1
   2. [ ] Step 2
   

3. •

   Run the audit using the CLI tool:

   bash

   python3 .agent/tools/utilities/gemini_audit.py --auto
   

   Or use the shell script:

   bash

   ./.agent/tools/bin/audit_task.sh
   

4. •

   Review the decision in the output:

   • •APPROVE: Plan aligns with task, proceed with implementation
   • •REJECT: Issues found, address before proceeding
   • •NEEDS_INFO: Clarification required, update task or plan
5. •

   The audit result is automatically appended to: .agent/docs/agent_handoffs/agent_conversation_log.md

Inputs and outputs

• •Inputs: .agent/task.md, .agent/implementation_plan.md
• •Outputs: Decision (APPROVE/REJECT/NEEDS_INFO), log entry appended

Constraints

• •Requires Gemini CLI initialized (run $gemini-cli-init first)
• •Auditor will not generate code, only critique
• •One audit per task/plan pair recommended before implementation

Examples

# Quick audit with defaults
python3 .agent/tools/utilities/gemini_audit.py --auto

# Explicit file paths
python3 .agent/tools/utilities/gemini_audit.py \
  --task .agent/task.md \
  --plan .agent/implementation_plan.md

# Shell script (uses defaults)
./.agent/tools/bin/audit_task.sh