{"id":50913077,"url":"https://github.com/garrus800-stack/garrus800-stack","last_synced_at":"2026-06-16T12:01:52.974Z","repository":{"id":349352502,"uuid":"1198848480","full_name":"Garrus800-stack/Garrus800-stack","owner":"Garrus800-stack","description":null,"archived":false,"fork":false,"pushed_at":"2026-05-13T22:06:22.000Z","size":46,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-05-14T00:30:46.549Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/Garrus800-stack.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-04-01T20:20:31.000Z","updated_at":"2026-05-13T22:06:26.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/Garrus800-stack/Garrus800-stack","commit_stats":null,"previous_names":["garrus800-stack/garrus800-stack"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/Garrus800-stack/Garrus800-stack","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Garrus800-stack%2FGarrus800-stack","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Garrus800-stack%2FGarrus800-stack/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Garrus800-stack%2FGarrus800-stack/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Garrus800-stack%2FGarrus800-stack/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Garrus800-stack","download_url":"https://codeload.github.com/Garrus800-stack/Garrus800-stack/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Garrus800-stack%2FGarrus800-stack/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34404748,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-16T02:00:06.860Z","response_time":126,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-06-16T12:01:51.073Z","updated_at":"2026-06-16T12:01:52.965Z","avatar_url":"https://github.com/Garrus800-stack.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"# Daniel — `@Garrus800-stack`\n\n**Solo architect of self-modifying AI systems.**\n\nBuilding agents that read their own code, fix their own bugs, and evolve their own architecture.\n\n![Profile views](https://komarev.com/ghpvc/?username=Garrus800-stack\u0026style=flat-square\u0026color=6c8cff\u0026label=profile+views)\n\n\u0026nbsp;\n\n[![Genesis Agent](https://img.shields.io/github/package-json/v/Garrus800-stack/genesis-agent?style=for-the-badge\u0026logo=electron\u0026logoColor=white\u0026color=6c8cff\u0026label=Genesis%20Agent)](https://github.com/Garrus800-stack/genesis-agent)\n[![Tests](https://img.shields.io/badge/tests-7937_passing-2da44e?style=for-the-badge)](https://github.com/Garrus800-stack/genesis-agent)\n[![Fitness](https://img.shields.io/badge/architectural_fitness-127%2F130-2da44e?style=for-the-badge)](https://github.com/Garrus800-stack/genesis-agent)\n[![Coverage](https://img.shields.io/badge/coverage-80%2F76%2F78-2da44e?style=for-the-badge)](https://github.com/Garrus800-stack/genesis-agent)\n\n---\n\n## Genesis Agent\n\nA self-aware, self-modifying cognitive AI agent with a 12-phase boot system, hexagonal architecture, and organism substrate. It doesn't just use LLMs — it wraps them in 376 modules of self-verification, self-repair, causal reasoning, autonomous planning, and runtime self-modification with rollback.\n\n```\n376 source files · ~119k LOC · 7937 tests · 178 services\n12 boot phases · 127/130 architectural fitness · coverage 80/76/78\n41 hash-locked safety files · 359 late-bindings · 18 CI audit gates\nCrash-safe sessions · Frontier-based memory · Heuristic self-scoring\nRuntime-toggleable subsystems · i18n EN/DE\nRuns on Claude, GPT-4, Qwen, DeepSeek, Kimi, or local models via Ollama\n```\n\n### 12-Phase Boot Architecture\n\n```mermaid\nflowchart TB\n    subgraph Top[Phase 9-12 — Agency Layer]\n        P9[Cognitive] --\u003e P10[Agency] --\u003e P11[Revolution] --\u003e P12[Late-Bind]\n    end\n    subgraph Mid[Phase 4-8 — Substrate]\n        P4[Hexagonal] --\u003e P5[Capabilities] --\u003e P6[Autonomy] --\u003e P7[Organism] --\u003e P8[Memory]\n    end\n    subgraph Bot[Phase 1-3 — Foundation]\n        P1[Foundation] --\u003e P2[Intelligence] --\u003e P3[Tools]\n    end\n    Bot --\u003e Mid --\u003e Top\n```\n\nBottom row resolves first — Container, EventBus, Settings, ModelBridge.\nMiddle row is the substrate — hexagonal ports, autonomous services, organism vitals.\nTop row is agency — GoalDriver, AgentLoop, cognitive workspace, frontier memory.\n359 late-bindings wire dependencies after all phases resolve.\n\n### Cognitive \u0026 Agency Subsystems\n\n```mermaid\nflowchart LR\n    AgentLoop --\u003e FormalPlanner\n    AgentLoop --\u003e Verifier\n    AgentLoop --\u003e WorldState\n    AgentLoop --\u003e EpisodicMemory\n    GoalDriver --\u003e AgentLoop\n    GoalDriver --\u003e GoalStack\n    AutonomousDaemon --\u003e GoalDriver\n    AutonomousDaemon --\u003e IdleMind\n    AutonomousDaemon --\u003e DesktopPerception\n    SelfModificationPipeline --\u003e AgentLoop\n    SelfModificationPipeline --\u003e SelfModVerification\n    SelfModificationPipeline --\u003e Reflector\n    Reflector --\u003e SnapshotManager\n    style AgentLoop fill:#6c8cff20\n    style GoalDriver fill:#6c8cff20\n    style AutonomousDaemon fill:#6c8cff20\n    style SelfModificationPipeline fill:#6c8cff20\n```\n\n### What Genesis Can Actually Do\n\n| Capability | What it Means |\n| --- | --- |\n| **Read own source code** | `read_source` and `mcp-code` tools — Genesis can introspect any module, including itself |\n| **Modify own code at runtime** | `SelfModificationPipeline` with hot-reload, sandbox testing, automatic rollback on regression |\n| **Detect own capability gaps** | `CapabilityHonesty` — refuses tasks it can't actually complete instead of hallucinating success |\n| **Plan multi-step goals** | `FormalPlanner` builds plans with verifiable steps; `GoalStack` for nested sub-goals; goal-relevant module paths injected into every plan-LLM prompt to prevent path hallucination |\n| **Verify own output** | `Verifier` runs post-execution checks; `SelfModVerification` ensures modifications don't regress |\n| **Recover from crashes** | `BootRecovery` with last-known-good snapshots; `SessionPersistence` for crash-safe context |\n| **Track own causality** | `CausalGraph` with edge confidence, contradiction detection, learned rules |\n| **Reason about own architecture** | `ArchitectureReflection` builds dependency graphs; `CognitiveSelfModel` for empirical self-awareness |\n| **Adapt prompts based on own performance** | `AdaptivePromptStrategy` rolls back regressions, promotes improvements via `PromptEvolution` A/B testing |\n| **Crystallize own skills** | `SkillCrystallizer` (Phase 2 Können) distills reusable skills from agent-loop trajectories; `SkillEffectivenessTracker` with Wilson scoring; `SkillPromotionEvaluator` gates which candidates graduate |\n\n### Safety Boundary\n\n| | |\n| --- | --- |\n| **SafeGuard** | Kernel + critical files hash-locked at boot (41 files protected) |\n| **CapabilityGuard** | Token-based scope enforcement |\n| **PreservationInvariants** | Hash-locked semantic safety rules |\n| **TrustLevelSystem** | Graduated autonomy (Supervised → Assisted → Autonomous → Full) |\n| **InjectionGate** | Defense against prompt injection in user input |\n| **DisclosurePolicy** | Trust-based information sovereignty |\n| **EmotionalState (read-only externally)** | No external override of emotional scalars |\n| **Self-Gate (observation-only)** | Genesis is never gated against thinking — only publishing decisions are gated |\n| **18 CI audit gates** | Hash-locked dev-time scripts that catch wiring drift, language discipline, doc-drift, service-number drift, listener leaks, raw timers, self-gate coverage, future-version references — all strict-mode |\n\n### Recent Milestones\n\n| Version | Highlight |\n| --- | --- |\n| **v7.9.6** | **Hygiene Pass + Pursuit-Loop Fixes + Path-Context Root** — toolRegistry→tools wiring fix, four missing `skill:*` channels into CHANNELS contract, two new audit gates (`audit-doc-language` + `audit-service-numbers`, 16→18), `lockCritical` 21→41 (CI scripts now hash-locked), three pursuit-loop fixes (final-return error field + replan-step normalization + hallucination regex extension), new shared `plan-context.js` helper feeds `pickRelevantModules` and \"use these EXACT paths\" directive into all three planners (`AgentLoopPlanner`, `FormalPlanner`, `ColonyOrchestrator`) closing the path-hallucination root, ARCHITECTURE +2 chapters (Hauptstandort/Außenposten identity topology + Self-Modification Pipeline as subsystem), English-only sweep in runtime logs |\n| **v7.9.5** | **Inhabit Activity + Architecture Graph Zoom + Doc Sweep** — 17th IdleMind activity (deterministic self-state snapshot via InnerSpeech, PSE HardGate-blocked); mouse-wheel zoom / click-drag pan / double-click reset on the architecture graph; live-fix pass (undo gated against gitAutoCommit+.git, shutdown summary 8s timeout, continuation maxAttempts setting, archreflect 15min + per-phase timing, daemon suggestions/health-issues persisted + slash, runtime-missing log, peer multicast honesty, lessonsAutoCapture diagnostic) |\n| **v7.9.4** | **SkillPromotionEvaluator + SkillRehearsal** — promotion gate decides which Phase-2-Können candidates graduate from rehearsal into active skills; rehearsal loop exercises candidates against historical contexts before promotion |\n| **v7.9.1** | **Live-Fix Pass** — `continue` auto-approved at AUTONOMOUS, ApprovalGate 5 min, `loop_early` filter, 24h goal-reject cooldown, IdleMind per-type activity counts, CHANGELOG englished, doc-counts reconciled |\n| **v7.9.0** | **Skill Forge + Phase 2 Können** — `SkillCrystallizer` distills reusable skills from agent-loop trajectories; `SkillEffectivenessTracker` with Wilson scoring; iteration loop in `SkillManager.createSkill`; format-tolerant `executeSkill`; `/run-skill` with JSON arg; `SkillRecallSection` in PromptBuilder |\n| **v7.8.9** | **Phase 1 Können — Affect-Encoding** — affect-trail snapshots at agent-loop boundaries with gate-pass status and per-loop pass-rate |\n| **v7.8.4** | **Bug-Sweep + Pre-Deletion Audit** — empty errorMessage fallback, StalledGoalWatchdog hardening, PathPlausibility, four-layer cleanup-protocol pattern |\n| **v7.7.9** | **Proactive Self-Expression Phase 3** — full PSE pipeline with all five `allowedKinds` active, first confirmed end-to-end PSE run |\n| **v7.7.8** | **Goal-Awareness** — five fixes (G1-G5) so Genesis perceives goal-state cleanly: closing-permission classifier, plan-issues never auto-approved, FormalPlanner canonical step types, conversational origins refused, plan-failure reflexion classified into five categories |\n| **v7.7.0–v7.7.7** | **Audit \u0026 Schema Hardening** — schema-store hardening, audit-doc-drift cross-checks, toolchain refresh, monaco ESM contract, dist-bundle stability, six audit findings closed |\n| **v7.6.0–v7.6.9** | **Track A Cleanup** — three structural splits (UI dual-path, CommandHandlersInstall, CommandHandlersOpen platform-resolver), eight-finding audit pass closed, fitness 130/130 |\n| **v7.5.9** | **Audit Closeout, Plan-Cards, Linux Readiness** — six audit bugs B1-B6 plus three rounds of Linux readiness fixes, Plan-Cards UI feature |\n| **v7.5.7** | **ModelRouter Settings UI** — two-column model-chain editor, default flip after CPU-Ollama timeout regression |\n| **v7.5.5** | **SelfStatementLog** — 789 LOC; auto-captures own statements, classifies into selbst-strukturell / selbst-emotional / selbst-versprechen |\n| **v7.5.0** | **ModelRouter-Autonomie** — `ModelBridge.chat()` queries router by taskType, switches internally; `agency.autoRouteByTask` setting |\n| **v7.4.7** | **Reinraum** — three dead settings made real, runtime toggles for Daemon / IdleMind / SelfMod, four new UI controls, i18n bridge for chat-system messages |\n| **v7.4.6** | **Goal-Pipeline Fixes** — first end-to-end Windows success on real goals; sandboxed shell execution with verbatim quoting and LLM-hallucination guards |\n| **v7.4.5** | **Durchhalten** — GoalDriver (P10) replaces frame-stack; CostStream + ResourceRegistry as P1 services; sub-goal spawn |\n| **v7.4.0–v7.4.4** | **Endurance \u0026 Cleanup** — CommandHandlers split (five focused modules from one monolith), bookkeeping passes, real answers over confident guesses, \"in the now\" — module ages out single-version pins |\n| **v7.3.7** | **Memory tools** — `mark-moment`, `journal-write`, `release-protected-memory` for Genesis to author its own episodic memory |\n| **v7.3.0** | **Hexagonal architecture** — Ports for LLM, ToolUse, Awareness, Effector, CodeSafety |\n| **v7.2.3** | **ONTOGENESIS** philosophy — code is habitat, `.genesis/` is identity. Updates are habitat-swaps, not replacements |\n| **v7.2.0** | Self-Define — Genesis writes its own identity from deterministic data |\n| **v7.1.5** | EmotionalFrontier — cross-layer emotional continuity. Inspired by [neo.mjs](https://github.com/neomjs/neo) Memory Core |\n| **v7.1.4** | Session-Aware Memory — crash-safe checkpoints, frontier node, cross-referencing |\n\n---\n\n## Solo Developer\n\nSolo developer. Every module has tests. Every event has a payload schema. Every safety claim is verified by a test that fails if the claim breaks. Architectural fitness is a CI gate. The codebase grows because it deserves to, not because the calendar says so.\n\nI plan before I code. I prefer one stable, meaningfully better release over five patched iterations. No marketing-driven version names — just bug-fix passes when that's what they are.\n\n**Stack:** JavaScript · Node.js · Electron · MCP Protocol · Ollama · Anthropic · OpenAI · DeepSeek · Kimi · Qwen\n\n---\n\n**Germany** · Building autonomous systems that understand their own architecture.\n\n[![Genesis Agent](https://img.shields.io/badge/%E2%86%92_Genesis_Agent-Main_Project-6c8cff?style=flat-square)](https://github.com/Garrus800-stack/genesis-agent)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fgarrus800-stack%2Fgarrus800-stack","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fgarrus800-stack%2Fgarrus800-stack","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fgarrus800-stack%2Fgarrus800-stack/lists"}