# MAP — Wo liegt was? Claude Code Analysen auf diesem System Stand: 2026-03-18 --- ## Verzeichnisbaum ``` /home/smlflg/ │ ├── Projekte/ │ ├── MetaMetaMeta/ ← DU BIST HIER (Konsolidierung) │ │ │ ├── Sidecar-ng/ │ │ ├── Claude_Code_Harness_Analyse_2026-03-16.md ★ KERN-DIAGNOSE │ │ │ → Attention Budget, additionalContext API, V8 Plan, A/B-Test │ │ ├── SIDECAR_EVOLUTION.md ★ VERSION HISTORY │ │ │ → V1-V7 Failure Patterns, warum NG funktioniert │ │ ├── ARCHITECTURE.md Sidecar-NG Architektur │ │ ├── REQUIREMENTS.md Funktionale/NF Specs │ │ ├── hook_inject.py (1172 LOC) UserPromptSubmit Hook │ │ ├── safety_gates.py (209 LOC) PreToolUse Gates │ │ ├── inject_log.jsonl Injection-Logs (Rohdaten) │ │ ├── safety_gates_log.jsonl Gate-Logs │ │ ├── eval_ab.py A/B Evaluation Framework │ │ ├── test_20_prompts.py Accuracy Tests │ │ ├── knowledge/ │ │ │ ├── CLAUDE.md Project-level Rules │ │ │ ├── WelcheFehlerVermeiden.md Top 10 Fehler │ │ │ ├── WieArbeitestDuMitSamuel.md Samuel-Profil │ │ │ ├── ADHD_TEMPLATE.md ADHS-Guardrails │ │ │ ├── skills-index.md Skill-Routing Index │ │ │ ├── Skilluebersicht.md Skill-Übersicht + Kosten │ │ │ ├── fear-map.md Angst-Trigger │ │ │ └── commands/ (34 Skills + gsd/ 36 Skills) Skill-Definitionen │ │ └── data/ │ │ └── skill_index.json (498 KB) 363 Open-Source Skills Index │ │ │ ├── Sidecar-V7-V2/ │ │ ├── FORENSIK.md ★ SCOPE CREEP FORENSIK │ │ │ → 83 Commits in 3 Tagen, Observer Paradox │ │ ├── HANDOFF.md Refactoring-Strategie (3863 LOC) │ │ ├── BERICHT.md Status-Bericht │ │ └── whattochange.md Änderungsliste │ │ │ ├── Sidecar-V8-Plan/ │ │ └── docs/ │ │ ├── architecture/ │ │ │ └── sidecar-v8-architecture.md ★ AGENT-AGNOSTIC DESIGN │ │ │ → Adapter-Pattern, Coupling Points, Multi-Agent │ │ └── evaluation/ │ │ └── gap-audit-2026-03-12.md ★ BUILD-BLOCKER │ │ → F1-F10 Gaps, M1-M7 Missing Decisions │ │ │ ├── Sidecar-V7/ │ │ ├── FOR_SMLFLG.md V7 Lessons, TTS, Speech Quality │ │ ├── PLAN_2026-03-13_ALIGNMENT_MONITOR.md 4 neue Mechanisms (#49-52) │ │ └── PLAN_2026-03-13_INJECTION_VISIBILITY.md additionalContext Visibility │ │ │ ├── Sidecar-Evolution/ ★ FEATURE ARCHAEOLOGY │ │ ├── Arbeitsanweisung/ │ │ │ ├── Eval-of-Eval-V1-V2.md Meta-Analyse: Eval-Prozess selbst │ │ │ ├── Eval-Review-V1.md 22 Features, 11 KEEP, 179 Tests │ │ │ ├── Eval-Review-V2.md 9 Features, 0 KEEP, 8 Duplikate │ │ │ ├── Eval-Review-V3.md LLM-Judge, bedingt-KEEP Pattern │ │ │ ├── Eval-Review-V4.md Knowledge-Base wins │ │ │ ├── Eval-Review-V5.md 75% KEEP-Rate (Höchste) │ │ │ ├── Evaluation_Methodik.md 7 Feature-Typen x 7 Eval-Strategien │ │ │ └── Sidecar_Evolution_Workflow.md Isolation-Protocol, Loop-Struktur │ │ ├── Cherry-Picks/ │ │ │ ├── V1-Enforcement-Rules.md 5 Workflow-Checks (56 LOC) │ │ │ ├── V1-Pattern-Detektoren.md 7 Detektoren (76 LOC, 71 Tests) │ │ │ ├── V1-Micro-Features.md 25 Features <20 LOC │ │ │ ├── V1-Safety-Gates.md 6 PreToolUse Gates │ │ │ └── park-list.md Features mit Fixes nötig │ │ ├── Kill-Liste/ │ │ │ └── kill-list.md 4 getötete Features │ │ └── NG/Dokumentation/OVERVIEW.md NG.1-NG.6 alle KEEP │ │ │ ├── ClaudeCodePanel/ │ │ ├── FOR_SMLFLG.md Panel Architecture, Quick Wins │ │ └── PLAN_2026-03-13_quickwins.md 9.7K LOC Refactoring │ │ │ ├── MyAIGame/ │ │ ├── CLAUDE.md Multi-Channel CLI, TTS Pipeline │ │ ├── claude-code/CLAUDE.md System Instructions Layer │ │ └── portable/codex/skills/ (30 Skills) Portable Skill-Kopien │ │ │ ├── samuel-os/ │ │ ├── ACTIVE.md Max 5 aktive Tasks │ │ ├── INBOX.md Triage Inbox │ │ └── DECISIONS.md Entscheidungs-Log │ │ │ ├── AgentArena/ │ │ ├── adapters/claude_code.py (209 LOC) CC CLI Benchmark-Adapter │ │ └── FOR_SMLFLG.md CC Score: 0.80 │ ├── AgentSwarms/ Swarm-Research │ ├── AgentSDK-HelloWorld/ │ │ └── FOR_SMLFLG.md OpenAI SDK statt Claude SDK (kein API Key) │ ├── Subconsciousness/ │ │ └── FOR_SMLFLG.md 3-Tier Memory, Hook-Integration │ └── (30+ weitere Projekte) │ ├── Dokumente/ │ ├── RETROSPEKTIVE_100_SESSIONS.md ★ SESSION META-ANALYSE │ │ → 60+ Sessions, ADHS-Patterns, Emotionale Landschaft │ │ │ ├── Ai_Cost_Engeneer/ │ │ └── WieVerwendeIchClaudeCode/ │ │ └── ANALYSE_KOMPLETT.md ★ SELF-AUDIT 6.1/10 │ │ → Kosten, Skills-Rationalisierung, 5 Sofort-Aktionen │ │ │ ├── Pläne/ │ │ ├── ClaudeCodeWorks/ │ │ │ ├── FOR_SMLFLG.md Audio/Multimodal via Hooks │ │ │ └── OPENCODETALKS.md OpenCode SSE-Bus Limitation │ │ ├── OpencodeWorks/ │ │ │ └── FOR_SMLFLG.md Template-Narration, Fallback Chain │ │ ├── CodexWorks/ Codex CLI Research │ │ ├── opencodethinks.md OpenCode Feature-Analyse │ │ ├── opencodeconclusion.md OpenCode Fazit │ │ ├── codexconclusion.md Codex Fazit │ │ ├── opencodeupdateforclaudecode.md Multi-Agent Hook-Integration │ │ ├── MultiKanalAgent.md Jarvis-Style Dual-Channel (Visual+Audio) │ │ ├── claudecodethinks.md Daemon vs Prompt-based Dual Output │ │ ├── SSE_TTS_Integration_Research.md SSE + Web Speech API │ │ └── PLAN_cross-session-fix-config-audit.md Config-Contamination Bug │ │ │ ├── Grosse_Fehler/ │ │ └── 001_sidecar_v7_zwei_tage.md ★ POST-MORTEM │ │ → 7 Blocker, Ghost-Processes, Rogue-Commit, Pipeline-Violation │ │ │ ├── Analysen/ │ │ ├── VERSION_SQUARED_FRAMEWORK.md ★ LEARNING EXTRACTION │ │ │ → L(n)=n², 140 Learnings V1-V7, Port:Fix 1:2.3 │ │ └── claude/ │ │ ├── session-analyse-sprint-session-2026-03-11.md │ │ └── version-archaeology-v1-v6.md │ │ │ ├── ClaudeAnalysiertMICH/ Claude's Analyse von Samuel │ ├── ADHS/ Persönliche ADHS-Dokumentation │ └── durchstarten/ $5.41 tracked │ ├── Downloads/ │ └── Claude T450s-20260209T160919Z-1-001/ │ └── Claude T450s/ ★ T450s ARCHIV (80+ Docs) │ ├── FINAL_SYNTHESIS_REPORT.md 74 Sessions, 3 Monate Meta │ ├── PRINCIPLES.md Lernphilosophie, 8-Wochen-Plan │ ├── ORCHESTRATION_GUIDE.md Multi-Agent Patterns, Kosten │ ├── BUILD_SPECS_INDEX.md 5 Skill-Templates │ ├── INTEGRATION_COMPLETE.md Chef-Opencode Integration │ ├── QUICK_START.md Unified Crew Workflow │ ├── VERTICAL_INTEGRATION.md Architecture Patterns │ ├── HIERARCHICAL_ORCHESTRATION.md Org Structure │ ├── CLAUDE.md Critical Rules (alt) │ ├── commands/ (crew, chef, langgraph, research) │ └── skills/ (documentation, refactoring, testing, architecture) │ ├── ClaudesReich/ ★ GROSSES ARCHIV │ ├── FOR_SMLFLG.md Archiv-Einführung │ ├── ACP_INFRASTRUCTURE_PLAN.md Agent Communication Protocol │ ├── claude-in-chrome-setup.md Chrome Integration Guide │ ├── ArchivPläne/ (48 Markdown Files) │ │ ├── EXECUTIVE_SUMMARY.md AI Testing ROI 500% │ │ ├── AI_TESTING_RESEARCH_SUMMARY.md (50 KB) ISTQB 2024, ACP, Frameworks │ │ ├── IMPLEMENTATION_TEMPLATES.md (28 KB) Production-Ready Code │ │ └── (44 weitere Session-Plans) │ ├── SessionBrowser/ ★ TOOL: 834 Sessions durchsuchbar │ │ ├── app.py Streamlit Web UI │ │ └── session_parser.py JSONL Parser + Cost Calculator │ ├── Visualisierung/ ★ TOOL: MAS Dashboard │ │ ├── app.py Streamlit Analytics │ │ └── views/ (warroom, timeline, sankey_costs, storyboard) │ ├── EveryClaudeCodeSessionfromeverydeviceever/ │ │ └── 3Laptops/ │ │ ├── Claude -insights/Gen16/ Gen16 Session Data (14 JSONL) │ │ └── Claude -insights/MacBookPro16/ MacBook Session Data │ └── speedread RSVP Speed-Reading Tool │ ├── ClaudeSystemBackup/ │ └── session-logs/ │ └── TAGESANALYSE_2026-03-09.md ★ HYPERFOCUS TAG │ → 10 Features, 0 E2E Tests, Verification 2/10 │ ├── Archiv/ │ └── SESSION_LOG.md Ältere Session-Logs │ ├── DataEngeeneeringKEEPA/ │ ├── BigAnalysis.md (560 Zeilen) Dead Code Analyse, LangGraph Decision │ └── KeepaProjectsforDataEngeneering3BranchesMerge/ │ ├── FOR_SMLFLG.md Opus für Planning, MiniMax für Dev │ └── docs/archive/ClaudeChromeExtension.md (358+) NativeMessaging Protocol │ ├── AiSystemForVibeCoding/ $39.74 tracked ├── Orderformyaistuff/ $4.61 tracked ├── Erdhügel/ $5.35 tracked ├── Moltbot/SelfAiMoltbot/skills/ Obsidian/Notion Skills ├── ai-coding-landscape-2026.html AI Coding Assistants Landscape ├── PLAN_2026-03-13_SWARM-AUTO-ATTACH.md Swarm UI Plan ├── Bilder/ CC Session Screenshots (02/2026) ├── claude-backup-2026-03-16.tar.zst (837 MB) Full Backup │ ├── .claude/ ★ KONFIGURATION │ ├── CLAUDE.md (16 KB, 280+ Zeilen) Global Master Instructions │ ├── settings.json (2.9 KB) Hooks, Permissions, Model │ ├── settings.local.json Local Overrides │ ├── agents/ (21 Agents) Agent-Definitionen │ ├── skills/eval/SKILL.md Eval Skill │ ├── hooks/ │ │ ├── usage-tracker.sh Tool Usage JSONL │ │ ├── event_logger.py Universal Event Log │ │ ├── gsd-context-monitor.js Context Warnings │ │ ├── gsd-check-update.js GSD Updates │ │ └── coaching/ │ │ ├── coach-before.sh Pre-Session (10 Rules, 40 Tips) │ │ ├── coach-after.sh Post-Session Compliance │ │ ├── coach-git-test.sh Git/Test Lessons │ │ └── coaching_log.md Session Reviews │ ├── session-logs/ (6 Files) Session-Log Archive │ ├── usage/ (Daily JSONL) Token Usage Tracking │ ├── events/ (Daily JSONL) Hook Event Logs │ ├── get-shit-done/ (Templates, Workflows) GSD Framework │ ├── projects/ (15+ Projekt-Kontexte) Session Memory │ ├── plans/ (30+ Files) Session-Plans │ ├── Symlinks → Sidecar-ng/knowledge/ │ │ ├── ADHD_TEMPLATE.md │ │ ├── commands/ │ │ ├── fear-map.md │ │ ├── Skilluebersicht.md │ │ ├── WelcheFehlerVermeiden.md │ │ └── WieArbeitestDuMitSamuel.md │ └── .credentials.json (600 perms) OAuth Tokens │ ├── .agent/ │ └── diagrams/ (28 HTML Files) ★ VISUALISIERUNGEN │ ├── claude-config-evolution.html (351 KB) │ ├── meta-analyse-4wochen-samuel.html │ ├── sidecar-ecosystem-overview.html │ ├── sidecar-ng-skill-router.html │ ├── claudecodepanel-deep-review.html │ ├── claude-directory-architecture.html │ └── (22 weitere) │ ├── .hermes/skills/ Hermes Agent Skills ├── .claude.json (109 KB) 534 Startups, Telemetrie └── .local/bin/ ├── claude-session-count.sh ├── claude-find-session ├── claude-restore └── claude-sessions ``` --- ## Legende ★ = Primäre Analyse-Dokumente (die wichtigsten 10) | # | Dokument | Pfad | Kernthema | |---|----------|------|-----------| | 1 | Harness-Analyse | Sidecar-ng/ | Attention Budget, additionalContext, V8 | | 2 | ANALYSE_KOMPLETT | Dokumente/Ai_Cost_Engeneer/ | Self-Audit 6.1/10, Kosten | | 3 | RETROSPEKTIVE_100 | Dokumente/ | 60+ Sessions, ADHS, Emotionen | | 4 | FORENSIK | Sidecar-V7-V2/ | Scope Creep, Observer Paradox | | 5 | TAGESANALYSE | ClaudeSystemBackup/ | Hyperfocus Tag, 0 Verification | | 6 | SIDECAR_EVOLUTION | Sidecar-ng/ | V1-V7 Failures, NG Design | | 7 | V8 Architecture | Sidecar-V8-Plan/ | Agent-Agnostic, Adapter | | 8 | V8 Gap-Audit | Sidecar-V8-Plan/ | Build-Blocker, Decisions | | 9 | FINAL_SYNTHESIS | Downloads/T450s/ | 74 Sessions, 3 Monate | | 10 | ORCHESTRATION_GUIDE | Downloads/T450s/ | Multi-Agent, Kosten | | 11 | VERSION_SQUARED | Dokumente/Analysen/ | L(n)=n², 140 Learnings, Port:Fix | | 12 | 001_sidecar_v7 | Dokumente/Grosse_Fehler/ | Post-Mortem, 7 Blocker | | 13 | Eval-of-Eval | Sidecar-Evolution/ | Meta-Analyse des Eval-Prozesses | | 14 | AI_TESTING_RESEARCH | ClaudesReich/ArchivPläne/ | ISTQB 2024, ACP, Frameworks | --- ## Nach Thema ### Diagnose & Probleme - Harness-Analyse → Attention Budget als Kernproblem - FORENSIK → Scope Creep Mechanik - TAGESANALYSE → Hyperfocus ohne Verification - RETROSPEKTIVE → ADHS-Patterns in der Nutzung - 001_sidecar_v7_zwei_tage → Post-Mortem mit 7 konkreten Blockern - VERSION_SQUARED_FRAMEWORK → Systematische Lern-Extraktion (140 Learnings) ### Self-Audits & Metriken - ANALYSE_KOMPLETT → 6.1/10, Top 5 Fehler, $148-200+ verschwendet - FINAL_SYNTHESIS → 74 Sessions, 27% Completion Rate - coaching_log.md → Agent Compliance, Streak Counter ### Architektur & Design - V8 Architecture → Agent-Agnostic Adapter-Pattern - SIDECAR_EVOLUTION → Warum Stateless > Stateful - HANDOFF → Refactoring 3863 LOC daemon.py ### Tool-Vergleiche - codexconclusion.md → Codex Features/Limits - opencodeconclusion.md → OpenCode Features/Limits - OPENCODETALKS.md → OpenCode SSE-Bus Problem - Harness-Analyse Section 2 → Hermes als Referenz ### Kosten & Optimierung - ORCHESTRATION_GUIDE → CrewAI $0.60 vs SubAgents $3 - ANALYSE_KOMPLETT → CLAUDE.md Bloat 42% Token-Reduktion - BUILD_SPECS_INDEX → $0.10-0.15 pro Skill-Template ### ADHS-spezifisch - ADHD_TEMPLATE.md → Guardrails - fear-map.md → Angst-Trigger - WieArbeitestDuMitSamuel.md → Profil - RETROSPEKTIVE Section "Emotional Landscape" ### Visualisierungen (HTML) - ~/.agent/diagrams/ → 28 interaktive Diagramme - claude-config-evolution.html (351 KB) ist das größte ### Feature Archaeology - Sidecar-Evolution/Arbeitsanweisung/ → V1-V5 Evaluierungen, Methodik - Cherry-Picks/ → Validierte Features mit Integration-Checklists - Kill-Liste/ → Getötete Features mit Begründung - Eval-of-Eval-V1-V2.md → Meta-Analyse des Eval-Prozesses selbst ### Post-Mortems & Fehler - Grosse_Fehler/001_sidecar_v7_zwei_tage.md → 7 Blocker, Ghost-Processes - ClaudeAnalysiertMICH/ → Claude's Selbst-Analyse ### Tools (selbst gebaut) - ClaudesReich/SessionBrowser/ → Streamlit App für 834 Sessions - ClaudesReich/Visualisierung/ → MAS Dashboard (War Room, Timeline, Sankey, Storyboard) - AgentArena/adapters/claude_code.py → Benchmark-Adapter (Score 0.80) ### Rohdaten & Logs - inject_log.jsonl → Jede Sidecar Injection - safety_gates_log.jsonl → Jede Gate Decision - ~/.claude/usage/*.jsonl → Token Costs pro Tag - ~/.claude/events/*.jsonl → Hook Events pro Tag - coaching_log.md → Session Reviews - ClaudesReich/EveryClaudeCodeSession.../ → JSONL Session-Daten von 3 Laptops --- ## Statistik | Metrik | Wert | |--------|------| | Claude Code Startups | 534 (seit 28.01.2026) | | Analysierte Sessions | ~100+ dokumentiert | | Primäre Analyse-Docs | 14 | | Sekundäre Docs | 50+ | | ArchivPläne (ClaudesReich) | 48 | | Session-Browser Sessions | 834 | | HTML Visualisierungen | 28 | | T450s Archiv Docs | 80+ | | Projekte in ~/Projekte/ | 38+ | | Konfigurierte Agents | 21 | | Skill-Definitionen | 69 (33 core + 36 GSD) | | Hook-Scripts aktiv | 7 | | Geschätzte Kosten analysiert | $100+ |