Update agents

[ymanor2404]

Temporarily moved agents to bullpen, enhanced 5 agent persona definitions that passed checks through Ernie the Evaluator Gem, and updated README.md

[github-actions]

Code Review: Agent Persona Updates

Summary

This PR contains significant updates to agent persona definitions, moving from concise bullet-point format to detailed, structured workflow descriptions. Additionally, several agents have been moved to an "agent-bullpen" folder to focus on 5 core agents for the current pilot.

---

✅ Strengths

1. More Actionable Agent Personas

The new format provides clear phase-by-phase workflows:

• Parker (PM): Excellent RICE framework integration with specific Jira workflow guidance
• Ryan (UX Researcher): Comprehensive research methodologies covering Discovery → Exploratory → Evaluative → Monitor phases
• Stella (Staff Engineer): Strong "multiplication" principle emphasizing impact through enabling teams

2. Industry-Standard Methodologies

• RICE scoring formula clearly documented for PM prioritization
• UX research phases match real-world practices (Field Studies, Card Sorts, Usability Testing, etc.)
• Staff Engineer role properly scoped as cross-team coordinator and technical authority

3. Clear Agent Reorganization Strategy

• Well-documented rationale in README for moving 12 agents to agent-bullpen/
• Maintains assets for future reintegration
• Focuses pilot on 5 core agents

---

⚠️ Issues Found

Critical (Must Fix)

1. Duplicate Content in parker-product_manager.md

Line 3: description: Product Manager Agent focused on market strategy...
Line 7: Description: Product Manager Agent focused on market strategy...

The description appears twice (frontmatter and body). Remove line 7.

High Priority

2. Lost Agent Personality/Voice
The old format had engaging "signature phrases" that gave each agent a distinct voice:

• Parker's: "Our customers are telling us..." and "How does this differentiate us from [competitors]?"
• Ryan's: "Our research shows that users actually..." and "Based on our usability testing..."
• Stella's: "Let me show you how we handled this in..." and "I'll pair with you on the tricky parts"

Recommendation: Add a brief "Voice & Communication Style" section at the top of each persona (100-150 words) before the detailed workflows.

3. Token Cost Impact

• Before: ~50 lines per agent × 5 = ~250 lines
• After: ~150 lines per agent × 5 = ~750 lines
• Impact: 3x increase in context usage per agentic session

While manageable with 200K context windows, this should be monitored. Consider:

• Adding session metrics to track context window usage
• Documenting the token cost trade-off in CLAUDE.md

Medium Priority

4. Inconsistent Formatting
Mixed heading styles across agent files:

# parker-product_manager.md
Line 9: "Personality & Communication Style (Retained & Reinforced)"  # Bold text, not header
Line 14: "Part 1: Define the Role's "Problem Space"..."          # Bold text, not header

Recommendation: Standardize on either markdown headers (##, ###) or bold text throughout.

5. No Schema Validation
Agent personas are free-form markdown with no validation. Consider:

• YAML frontmatter schema validation
• Required fields check (name, description, tools, phases)
• Automated tests in CI

6. Missing Tests
No test coverage for:

• Agent persona file loading/parsing
• Schema completeness validation
• Integration with Claude Code runner

---

📊 Detailed File Reviews

agents/parker-product_manager.md ⭐ EXCELLENT

• Well-structured 4-phase workflow (Opportunity → Prioritization → Definition → Launch)
• RICE formula clearly explained: (Reach × Impact × Confidence) / Effort
• Specific Jira workflow integration details
• Issue: Duplicate description (line 3 & 7)

agents/ryan-ux_researcher.md ⭐ EXCELLENT

• Comprehensive research methodology coverage
• Clear distinction between generative vs evaluative research
• Specific study types with definitions (Field Study, Diary Study, Card Sort, Tree Test, etc.)
• Practical outputs for each phase (Empathy Maps, Journey Maps, Usability Issues by Severity)
• Strength: Reads like a professional UX research handbook

agents/stella-staff_engineer.md ⭐ GOOD

• Strong "multiplication" core principle (impact through enabling others)
• Clear 4-phase structure aligned with product lifecycle
• Good emphasis on cross-team coordination and mentorship
• Strength: Properly scopes Staff Engineer as technical leader vs individual contributor

agents/steve-ux_designer.md & terry-technical_writer.md

(Not shown in diff excerpt, but same structural approach applies)

---

🔒 Security & Best Practices

✅ No Security Issues Found

• No credentials, tokens, or sensitive data
• No code execution, only documentation changes
• Follows repository markdown conventions

✅ Follows CLAUDE.md Standards

• Proper frontmatter with name, description, tools
• Markdown format consistent with project docs

---

📋 Recommendations Summary

Must Do Before Merge:

1. ✅ Remove duplicate description in parker-product_manager.md (line 7)

Strongly Recommended:

2. Add brief "Voice & Personality" summaries (100-150 words) at the top of each agent
3. Document token cost trade-offs in CLAUDE.md or README
4. Verify no code references moved agents: git grep -r "agents/archie" "agents/aria" "agents/casey"

Future Improvements:

5. Add YAML schema validation for agent frontmatter
6. Create automated tests for agent persona loading
7. Standardize formatting (headers vs bold text)
8. Add agent-to-workflow mapping table in README

---

✅ Overall Assessment: APPROVE with Minor Changes

This is a valuable improvement that makes agent personas significantly more actionable for Claude Code. The structured workflow approach with specific methodologies, deliverables, and key questions will help in complex multi-agent sessions.

Recommended Action:

1. Fix the duplicate description in Parker's file
2. Consider adding personality summaries back
3. Merge and monitor session performance metrics

Key Trade-off:
You've exchanged concise, voice-heavy personas for detailed, workflow-heavy references. This is the right choice for a pilot focused on structured deliverables, but be prepared to optimize if context window usage becomes an issue.

---

Great work on the restructuring! The industry-standard methodologies (RICE, UX research phases) and clear deliverables make these agents much more practical for real product development workflows. 🚀

[Gkrumbach07]

Looks good. I will merge this but can you also do the same in this repo now https://github.com/ambient-code/ootb-ambient-workflows

[Gkrumbach07]

Tracked in Jira: https://issues.redhat.com/browse/RHOAIENG-37656