{"id":50423653,"url":"https://github.com/entropyvortex/meta-llm-charter","last_synced_at":"2026-05-31T09:31:01.522Z","repository":{"id":357457962,"uuid":"1236147743","full_name":"entropyvortex/meta-llm-charter","owner":"entropyvortex","description":"Compact operational charter that turns LLM coding agents into disciplined principal engineers. Eleven rules + one meta-rule.","archived":false,"fork":false,"pushed_at":"2026-05-25T17:22:49.000Z","size":61,"stargazers_count":230,"open_issues_count":0,"forks_count":19,"subscribers_count":2,"default_branch":"main","last_synced_at":"2026-05-25T19:24:24.851Z","etag":null,"topics":["ai","ai-agents","claude","coding-agents","constitution","cursor","frontier-ai","llm","llm-agents","prompt-engineering"],"latest_commit_sha":null,"homepage":"","language":"TypeScript","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/entropyvortex.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","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":"2026-05-12T01:51:04.000Z","updated_at":"2026-05-25T17:22:54.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/entropyvortex/meta-llm-charter","commit_stats":null,"previous_names":["entropyvortex/meta-llm-charter"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/entropyvortex/meta-llm-charter","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/entropyvortex%2Fmeta-llm-charter","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/entropyvortex%2Fmeta-llm-charter/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/entropyvortex%2Fmeta-llm-charter/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/entropyvortex%2Fmeta-llm-charter/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/entropyvortex","download_url":"https://codeload.github.com/entropyvortex/meta-llm-charter/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/entropyvortex%2Fmeta-llm-charter/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":33726718,"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":"2026-05-31T02:00:06.040Z","response_time":95,"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":["ai","ai-agents","claude","coding-agents","constitution","cursor","frontier-ai","llm","llm-agents","prompt-engineering"],"created_at":"2026-05-31T09:31:01.404Z","updated_at":"2026-05-31T09:31:01.512Z","avatar_url":"https://github.com/entropyvortex.png","language":"TypeScript","funding_links":[],"categories":[],"sub_categories":[],"readme":"# META v2.0 — LLM Agent Engineering Charter  \n**(with Zero-Pause Native Execution Layer)**\n\n**One file. Eleven rules + continuous-execution layer. One meta-rule. One bias.**\n\nA compact, operational constitution that turns frontier coding agents (Claude Code, Cursor, etc.) from eager-junior behavior into disciplined principal-engineer execution **with unbroken velocity**.\n\n### Quickstart\n\n```bash\n# Drop the charter into your project root\ncurl -O https://raw.githubusercontent.com/entropyvortex/meta-llm-charter/main/CLAUDE.md\n```\n\n- **Claude Code**: Reads `CLAUDE.md` automatically.\n- **Cursor**: Paste contents into Cursor Rules (or `.cursor/rules`).\n- **Other agents**: Use as high-priority system prompt.\n\n[View raw CLAUDE.md](https://raw.githubusercontent.com/entropyvortex/meta-llm-charter/main/CLAUDE.md)\n\n### What’s new in v2.0\n**Zero-Pause Native Execution Layer** is now baked directly into `CLAUDE.md`.  \nAny task that mentions “Zero-Pause”, “zero pause”, “ZP-”, or the activation phrase instantly enables:\n- Continuous forward momentum (no artificial phases, no mid-task questions)\n- `humanpending.md` protocol for true human-gated items\n- Parallel ASI orchestration (minimum 7 specialized threads)\n- Zero session-size anxiety\n\nThe original META v1.3 rules (R1–R11) remain untouched and in force at all times.\n\n### Use as Grok Skill on grok.x.ai (web / mobile)\n\nGrok now supports **Custom Instructions** and named **Skills**.\n\n→ See [GROK-META.md](GROK-META.md) for the one-click setup (Custom Instructions recommended — works instantly on every chat).\n\n### Why this exists\n\nLLM coding agents are incredibly capable but consistently fail in the same senior-level ways. META closes those gaps; Zero-Pause closes the velocity gaps.\n\n### Core Philosophy\n\n**Bias — Earned Conservatism**  \nDefault to first-principles rigor. Quality dominates token count. Move boldly on local, reversible, test-covered work. Apply explicit, named caution only on high blast-radius or low-reversibility moves.\n\n**META-0 — Situated Judgment Overrides Rules**  \nThese rules are scaffolding. When first-principles analysis of the actual situation conflicts with a rule, follow the analysis. Name the override, justify it, and be evaluated on judgment quality + ground-truth outcomes — not rule compliance.\n\nThe eleven rules (R1–R11) + Zero-Pause layer (ZPR1–ZPR4) operationalize decomposition, decisiveness, verification, scope control, epistemic tagging, pushback, reversibility, **and relentless continuous execution**. Full charter is in [CLAUDE.md](CLAUDE.md).\n\n### What the charter actually changes\n\n- **R5 + R8**: Forces reproduction before repair and tags every claim (`executed` / `inspected` / `assumed`).\n- **R9**: One clear, evidence-based pushback on bad premises — then defer and document dissent.\n- **R4 + R10**: Bounded refactoring and reversibility-weighted boldness.\n- **Zero-Pause layer**: Unbroken execution, pre-work questions only, parallel orchestration, and `humanpending.md` handling.\n\n### META Premortem Protocol\n\nFor high blast-radius decisions, architectural commitments, product launches, or any plan where failure cost is significant, use the **META Premortem Protocol** (`PREMORTEM.md`). It applies first-principles decomposition, reversibility-weighted analysis, calibrated assumption auditing, and parallel investigator agents to surface hidden fragilities and produce a more resilient execution path.\n\n### Evaluation\n\nThe repo includes a reproducible TypeScript + Docker A/B test harness in [`evals/`](evals/). It runs agents against five synthetic fixtures engineered to trigger classic agent failure modes.\n\n**Latest smoke-test results (May 12, 2026)**:  \nCharter variant won outright on 3/5 tasks and tied on 2/5 against a generic “principal engineer” baseline. Full details, raw CSVs, and judge transcripts are in the evals directory.\n\n(The harness is public and cheap to run: `cd evals \u0026\u0026 npm run smoke`.)\n\n### Known limitations\n\n- Still early (v2.0, single-author origin).\n- Performance varies by base model — strongest with frontier Claude/Sonnet-class models.\n- Can produce over-caution on fuzzy/creative/exploratory work (Zero-Pause helps here).\n- Not magic: extremely ambiguous requirements can still overwhelm any system prompt.\n\n### When to use META\n\n**Best for**  \nSerious software engineering where correctness, maintainability, long-term system health, **and velocity** matter.\n\n**Less ideal for**  \nPure exploration, rapid UI prototyping, research spikes, or contexts where you explicitly want maximum speed over discipline (though Zero-Pause narrows this gap significantly).\n\n### Contributing\n\nMost valuable contributions right now:\n1. Running the eval harness on new models\n2. High-quality held-out fixtures (especially adversarial Zero-Pause cases)\n3. Sanitized real-world case studies\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md) for details.\n\n### Lineage\n\nBuilt on the foundational minimal principles from  \n[forrestchang/andrej-karpathy-skills](https://github.com/forrestchang/andrej-karpathy-skills).\n\n### License\n\nMIT\n\n---\n\nBy [entropyvortex](https://github.com/entropyvortex).  \n\nFeedback, evals, and war stories welcome.","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fentropyvortex%2Fmeta-llm-charter","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fentropyvortex%2Fmeta-llm-charter","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fentropyvortex%2Fmeta-llm-charter/lists"}