refactor(humaninloop): recreate plan-architect for AGENT-GUIDELINES compliance

[deepeshBodh]

Summary

• Remove all five anti-leak violations from the plan-architect agent (phase branching, hardcoded paths, sibling awareness, context schema knowledge, sequencing knowledge)
• Rebuild as a pure persona agent with zero workflow coupling per AGENT-GUIDELINES.md
• Preserve core identity, skills, and capabilities while reducing from ~1,200 words (process-heavy) to ~380 words (pure persona)

What Changed

Dimension	Before	After
Phase branching	3 phase sections (Research, Data Model, Contracts)	Removed — skills handle activities
Hardcoded paths	.workflow/planner-report.md, research.md, etc.	"Write outputs to locations specified in your instructions"
Sibling awareness	"Ready for Devil's Advocate review"	Removed
Context schema	5-field context file definition	Removed
Sequencing knowledge	Phase ordering, "from phase 1"	Removed
Report format	Embedded template	Removed — belongs in skill/supervisor
Brownfield sections	2 sections with hardcoded paths	Removed — context comes from instructions
Description	No example blocks	3 example blocks with triggering conditions

Audit Result

Passed agent-auditor with verdict: PASS (0 Critical, 0 Important, 1 Minor note about word count being naturally concise due to clean skill delegation)

Test plan

• Verify agent invokes patterns-technical-decisions skill when given a research task
• Verify agent invokes patterns-entity-modeling skill when given a data modeling task
• Verify agent invokes patterns-api-contracts skill when given an API design task
• Verify agent maintains persona under pressure (rejects shallow research, demands traceability)
• Verify agent works in /humaninloop:plan workflow without coupling issues

🤖 Generated with Claude Code