refactor(humaninloop): bring plan-architect to AGENT-GUIDELINES compliance

[deepeshBodh]

Summary

• Strip all workflow-coupled content from plan-architect.md (256 → 102 lines, 70 additions / 224 deletions)
• Remove 5 anti-leak violations: phase branching, 10+ hardcoded paths, sibling agent reference (Devil's Advocate), context file schema, sequencing knowledge
• Remove 4 forbidden sections: Phase Behaviors, Report Format, How You Operate, Reading the Context
• Add 3 <example> blocks to description for proper triggering
• Add missing "What You Produce" section listing artifact types
• Strengthen Core Identity with specific experiential beliefs that shape concrete judgments
• Fold brownfield awareness into Core Identity and What You Embrace as character traits
• Use namespace syntax (humaninloop:patterns-*) for all skill cross-references
• Expand Quality Standards with architectural specifics (bidirectional relationships, state machine completeness, PII marking, impedance mismatches)

Audit result: PASS WITH NOTES (from REJECT with 11 critical issues)

Test plan

• Verify skill delegation: give agent a planning task and confirm it invokes humaninloop:patterns-technical-decisions rather than improvising
• Verify persona fidelity: give agent a task tempting single-option decisions and confirm it rejects
• Verify workflow independence: invoke from a supervisor prompt and confirm no coupling leaks

🤖 Generated with Claude Code