4.8 KiB

Raw Blame History

name	description	workflow_path	thisStepFile	nextStepFile
step-04-adversarial-review	Context-independent adversarial diff review via subagent - no story knowledge	{project-root}/_bmad/bmm/workflows/4-implementation/code-review	{workflow_path}/steps/step-04-adversarial-review.md	{workflow_path}/steps/step-05-consolidate-findings.md

Step 4: Adversarial Review (Information Asymmetric)

Goal: Perform context-independent adversarial review of code changes. Reviewer sees ONLY the diff - no story, no ACs, no context about WHY changes were made.

Reviewer has FULL repo access but NO knowledge of WHY changes were made DO NOT include story file in prompt - asymmetry is about intent, not visibility This catches issues a fresh reviewer would find that story-biased review might miss

AVAILABLE STATE

From previous steps:

{story_path}, {story_key}
{git_changed_files} - Files changed according to git
{context_aware_findings} - Findings from Phase 1

STATE VARIABLE (capture now)

{diff_output} - Complete diff of changes
{asymmetric_findings} - Findings from adversarial review

EXECUTION SEQUENCE

1. Construct Diff

Build complete diff of all changes for this story.

Determine diff source:

If uncommitted changes exist for story files:

git diff
git diff --cached

If story work is already committed, find story-related commits:

# Find commits that reference this story
git log --oneline --all --grep="{story_key}" --format="%H"
# Or find recent commits touching story files
git log --oneline -10 -- {story_file_list}

Then construct diff:

git diff {first_story_commit}^..HEAD -- {files}

Include in {diff_output}:

All modified tracked files (except files in {implementation_artifacts} - asymmetry requires hiding intent)
All new files created for this story
Full content for new files

2. Invoke Adversarial Review

Use information asymmetry: separate context from review

Execution Hierarchy (try in order):

Option A: Subagent (Preferred)

If Task tool available with subagent capability:

<invoke-task subagent="true">
  Review {diff_output} using {project-root}/_bmad/core/tasks/review-adversarial-general.xml
</invoke-task>

The subagent:

Has FULL read access to the repository
Receives ONLY {diff_output} as context
Does NOT know story requirements, ACs, or intent
Reviews code purely on technical merit

Option B: CLI Fallback

If subagent not available but CLI available:

# Pipe diff to adversarial review task
cat {diff_file} | claude --task {adversarial_review_task}

Option C: Inline Execution

If neither available, load review-adversarial-general.xml and execute inline:

Load task file
Adopt adversarial persona
Review {diff_output} with zero story context
Generate findings

3. Process Adversarial Findings

Capture findings from adversarial review.

If zero findings returned:

HALT - Zero findings is suspicious. Re-analyze or ask for guidance.

For each finding:

Assign severity:

CRITICAL: Security vulnerabilities, data loss risks
HIGH: Logic errors, missing error handling
MEDIUM: Performance issues, code smells
LOW: Style, documentation

Assign validity:

REAL: Genuine issue to address
NOISE: False positive (explain why)
UNDECIDED: Needs human judgment

Create {asymmetric_findings} list:

{
  id: "AAF-{n}",
  source: "adversarial-review",
  severity: "...",
  validity: "...",
  description: "...",
  location: "file:line (if applicable)"
}

4. Phase 2 Summary

Present adversarial findings:

**Phase 2: Adversarial Review Complete**

**Reviewer Context:** Pure diff review (no story knowledge)
**Findings:** {count}
- CRITICAL: {count}
- HIGH: {count}
- MEDIUM: {count}
- LOW: {count}

**Validity Assessment:**
- Real issues: {count}
- Noise/false positives: {count}
- Needs judgment: {count}

Proceeding to findings consolidation...

NEXT STEP DIRECTIVE

CRITICAL: When this step completes, explicitly state:

"NEXT: Loading step-05-consolidate-findings.md"

SUCCESS METRICS

Diff constructed from correct source (uncommitted or commits)
Story file excluded from diff
Subagent invoked with proper isolation (or fallback used)
Adversarial review executed
Findings captured with severity and validity
{asymmetric_findings} populated
Phase summary presented
Explicit NEXT directive provided

FAILURE MODES

Including story file in diff (breaks asymmetry)
Skipping adversarial review entirely
Accepting zero findings without halt
Not using subagent when available
Missing severity/validity classification
Not storing findings for consolidation
No explicit NEXT directive at step completion

4.8 KiB Raw Blame History