Weyermann vs Oracle System Analysis - Dec 9, 2025

2025-12-09_weyermann-vs-oracle-analysis.md

System Comparison: Weyermann vs Oracle

Date: 2025-12-09 Purpose: Human review before implementation Status: PENDING REVIEW

---

Quick Summary

What	Weyermann	Oracle	Action
CLAUDE.md	5 modular files (740 lines)	None	CREATE
Commands	15	2	COPY 2 more
Learnings	12 files	2 files	OK (same pattern)
Sub-agents	9	2	OK for now

---

1. CLAUDE.md Architecture

Weyermann (Complex)

claude.md (hub ~150 lines)
├── CLAUDE_safety.md (62 lines)
├── CLAUDE_lessons.md (99 lines)
├── CLAUDE_subagents.md (95 lines)
├── CLAUDE_workflows.md (149 lines)
└── CLAUDE_templates.md (335 lines)

Features:

• Navigation breadcrumbs in every file
• Priority reading levels (Required/As needed/Reference)
• Golden Rules (5 safety rules)
• Short codes quick reference

Oracle (Current)

• No CLAUDE.md exists

Recommendation

Create simple single CLAUDE.md first, not modular. Contains:

• Project purpose (Shadow/Oracle concept)
• Golden rules (context-finder first, etc.)
• Sub-agents list
• Commands list
• Folder structure

Complexity level: Start simple, modularize later if needed

---

2. Commands Comparison

Command	Weyermann	Oracle	Purpose
/rrr	✅	✅	Retrospective
/recap	✅	✅	Fresh start summary
/snapshot	✅	❌	Quick learning → local file
/ccc	✅	❌	Context → GitHub issue
/distill	✅	❌	Summarize content
speckit.*	✅ (9 cmds)	❌	Spec management

What to Copy

Priority 1: /snapshot

• Quick knowledge capture
• No GitHub issue (local only)
• Output: ψ-logs/YYYY-MM/DD/HH.MM_title.md
• Use case: "Just learned something, save it fast"

Priority 2: /ccc

• Context preservation before compact
• Creates GitHub issue with state
• Use case: "About to run out of context, save state"

Not needed yet: speckit.* (complex, for spec management)

---

3. Learnings Structure

Weyermann

ψ-learnings/
├── 001-force-push.md
├── 002-golden-rule.md
├── 003-upstream-first.md
├── 004-psi-naming.md
├── 005-pr-workflow-discipline.md
├── 006-gogogo-delegation-workflow.md
├── consult-before-acting.md
├── reference-based-cleanup.md
├── subagent-prompt-engineering.md
└── weekly/2025-w49.md

Oracle

learnings/
├── 001-delegate-reading.md
├── 002-context-finder-first.md

Analysis

• Same numbering pattern (001-NNN)
• Oracle doesn't use ψ- prefix (simpler)
• Weekly summaries could be useful later

Verdict: Oracle's approach is good. No change needed.

---

4. Folder Structure Comparison

Weyermann

ψ-learnings/     → Meta knowledge
ψ-logs/          → Snapshots
ψ-retrospectives/→ Session reviews
ψ-archive/       → Old stuff
ψ-docs/          → Documentation

Oracle

learnings/       → Meta knowledge
journal/         → Daily logs
retrospectives/  → Session reviews
context/         → Analysis files
raw/             → Ground truth data
ideas/           → Incubation
templates/       → Response templates
evolution/       → CLAUDE.md snapshots

Analysis

• Oracle has MORE organized structure
• context/ and raw/ are unique to Oracle (workshop data)
• ideas/ for incubation is good
• No ψ- prefix = simpler

Verdict: Oracle's structure is better for its purpose (Shadow/Oracle concept)

---

5. Sub-agents Comparison

Weyermann (9 agents)

Agent	Model	Purpose
context-finder	haiku	Search git/issues
archiver	haiku	Archive unused
coder	opus	Code creation
executor	haiku	Run from issues
issues-cleanup	haiku	GitHub management
marie-kondo	haiku	File placement
maw	haiku	Multi-agent workflow
new-feature	haiku	Plan features
thai-translator	haiku	Translate

Oracle (2 agents)

Agent	Model	Purpose
context-finder	haiku	Search git/issues
security-scanner	haiku	Pre-commit security

Analysis

• security-scanner is unique to Oracle (good!)
• Most weyermann agents are for multi-agent workflow (MAW)
• Oracle doesn't need MAW agents yet

Verdict: Add agents as needed, not all at once

---

6. What Makes Weyermann Special

Short Codes (Workflow Shortcuts)

nnn     → Create plan issue
gogogo  → Execute plan
rrr     → Retrospective
ccc     → Context capture

These create a conversation flow:

1. nnn — Plan what to do
2. gogogo — Do it
3. rrr — Reflect on what happened
4. ccc — Save context before compact

Self-Improving Loop

1. MEASURE → Score current state
2. IDENTIFY → Find problem
3. REFINE → Fix one thing
4. TEST → Verify
5. ITERATE → Repeat

Delegation Model

Main (Opus) → orchestrates
Sub-agents (Haiku) → execute
Result: 10x cheaper, cleaner context

---

7. Recommended Actions

Do Now

• Create simple CLAUDE.md
• Copy /snapshot command
• Copy /ccc command

Do Later

• Add weekly learning summaries
• Consider nnn and gogogo commands
• Add more sub-agents if needed

Don't Do

• Modular CLAUDE.md (too complex for now)
• Copy all 9 sub-agents (not needed)
• Use ψ- prefix (keep it simple)

---

8. Human Review Questions

Before implementing, please confirm:

1. CLAUDE.md: Single file or modular?

   • Recommend: Single file first

2. Commands: Copy /snapshot and /ccc?

   • Recommend: Yes, both useful

3. Logs folder: Create logs/ for snapshots?

   • Recommend: Yes, for /snapshot output

4. Naming: Keep no-prefix or adopt ψ-?

   • Recommend: Keep no-prefix (simpler)

---

Analysis by context-finder (parallel queries) + main agent synthesis Ready for human review