DISCOVERIES Archive¶
Archived: 2025-11-25
This file contains historical discoveries that have been resolved, superseded, or are no longer actively referenced. They're preserved here for historical context and git archaeology.
To restore an entry: If a pattern re-emerges or becomes relevant again, move it back to DISCOVERIES.md.
Table of Contents¶
- Project Initialization (2025-01-16)
- Anti-Sycophancy Guidelines (2025-01-17)
- Enhanced Agent Delegation (2025-01-17)
- Agent Priority Hierarchy Flaw (2025-01-23)
- Pre-commit Hooks Over-Engineering (2025-09-17)
- CI Failure Resolution Process (2025-09-17)
- Context Preservation Success (2025-09-23)
- Reflection System Data Flow Fix (2025-09-26)
- Claude-Trace UVX Argument Issue (2025-09-26)
Project Initialization (2025-01-16)¶
Status: Historical - foundational context
Issue¶
Setting up the agentic coding framework with proper structure and philosophy.
Solution¶
Created comprehensive .claude directory structure with context files, agent definitions, command system, hook system, and runtime directories.
Key Learnings¶
- Structure enables AI effectiveness
- Philosophy guides decisions
- Patterns prevent wheel reinvention
- Agent specialization works
Anti-Sycophancy Guidelines Implementation (2025-01-17)¶
Status: Resolved - now codified in TRUST.md
Issue¶
Sycophantic AI behavior erodes user trust.
Solution¶
Created TRUST.md with 7 anti-sycophancy rules. Added to standard imports in CLAUDE.md.
Key Learnings¶
Trust comes from honesty, not harmony. Directness builds credibility.
Enhanced Agent Delegation Instructions (2025-01-17)¶
Status: Resolved - now in CLAUDE.md
Issue¶
CLAUDE.md had minimal guidance on when to use specialized agents.
Solution¶
Added "GOLDEN RULE" emphasizing orchestration, specific delegation triggers for all agents, parallel execution examples.
Key Learnings¶
Explicit triggers drive usage. Orchestration mindset matters. Documentation drives behavior.
Agent Priority Hierarchy Critical Flaw (2025-01-23)¶
Status: Resolved - USER_REQUIREMENT_PRIORITY.md created
Issue¶
Agents were overriding explicit user requirements in favor of philosophy.
Solution¶
Created USER_REQUIREMENT_PRIORITY.md with mandatory hierarchy. Updated critical agents with requirement preservation checks.
Key Learnings¶
User explicit requirements are sacred. Philosophy guides HOW, not WHAT.
Pre-commit Hooks Over-Engineering (2025-09-17)¶
Status: Resolved
Issue¶
Initial pre-commit had 11+ hooks and 5 config files, violating ruthless simplicity.
Solution¶
Reduced to 7 essential hooks. Deleted custom philosophy checker, detect-secrets, complex pytest hook.
Key Learnings¶
Start minimal, grow as needed. Philosophy enforcement belongs in review, not automation.
CI Failure Resolution Process Analysis (2025-09-17)¶
Status: Lessons captured, aged out
Issue¶
Complex CI debugging took 45 minutes due to version mismatches and merge conflicts.
Solution¶
Multi-agent orchestration approach. Identified CI Version Mismatch and Silent Pre-commit Hook Failure patterns.
Key Learnings¶
Agent orchestration works for debugging. Environment parity is critical. Pattern recognition accelerates resolution.
Context Preservation Implementation Success (2025-09-23)¶
Status: Resolved - system implemented
Issue¶
Original user requests lost during context compaction.
Solution¶
Four-component system: Context Preservation, Enhanced Session Start Hook, PreCompact Hook, Transcript Management.
Key Learnings¶
Proactive preservation beats reactive recovery. Session-level context injection is most effective.
Reflection System Data Flow Fix (2025-09-26)¶
Status: Resolved
Issue¶
AI-powered reflection failing silently with "No session messages found".
Solution¶
Fixed interface contract - stop.py now passes messages directly to reflection.py.
Key Learnings¶
Data flow mismatches between legacy and new systems cause silent failures.
Claude-Trace UVX Argument Passthrough Issue (2025-09-26)¶
Status: Resolved
Issue¶
UVX argument passthrough failing for claude-trace integration.
Solution¶
Modified build_claude_command() in launcher/core.py to handle claude-trace's --run-with chat requirement.
Key Learnings¶
Different tools require different argument structures. Test real deployment scenarios.
End of Archive¶
For current discoveries, see DISCOVERIES.md.