{"id":48898508,"url":"https://github.com/arcobaleno64/consilium-fabri","last_synced_at":"2026-04-16T13:00:54.171Z","repository":{"id":350361273,"uuid":"1205745580","full_name":"arcobaleno64/consilium-fabri","owner":"arcobaleno64","description":"Consilium Fabri — Gate-guarded, artifact-first workflow template for multi-agent AI development (Claude Code + Gemini CLI + Codex CLI)","archived":false,"fork":false,"pushed_at":"2026-04-10T01:47:54.000Z","size":51,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2026-04-10T03:14:51.190Z","etag":null,"topics":["ai-orchestration","artifact-first","claude-code","developer-tools","gate-guarded","gemini-cli","harness-engineering","multi-agent","premortem","workflow-template"],"latest_commit_sha":null,"homepage":"","language":"Python","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/arcobaleno64.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"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":"AGENTS.md","dco":null,"cla":null}},"created_at":"2026-04-09T08:40:35.000Z","updated_at":"2026-04-10T01:47:59.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/arcobaleno64/consilium-fabri","commit_stats":null,"previous_names":["arcobaleno64/consilium-fabri"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/arcobaleno64/consilium-fabri","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/arcobaleno64%2Fconsilium-fabri","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/arcobaleno64%2Fconsilium-fabri/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/arcobaleno64%2Fconsilium-fabri/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/arcobaleno64%2Fconsilium-fabri/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/arcobaleno64","download_url":"https://codeload.github.com/arcobaleno64/consilium-fabri/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/arcobaleno64%2Fconsilium-fabri/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":31886937,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-04-16T11:36:10.202Z","status":"ssl_error","status_checked_at":"2026-04-16T11:36:09.652Z","response_time":69,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.5:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"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-orchestration","artifact-first","claude-code","developer-tools","gate-guarded","gemini-cli","harness-engineering","multi-agent","premortem","workflow-template"],"created_at":"2026-04-16T13:00:27.420Z","updated_at":"2026-04-16T13:00:54.164Z","avatar_url":"https://github.com/arcobaleno64.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\n# Consilium Fabri\n\n\u003cp\u003e\n  A production-minded multi-agent AI workflow for teams that want traceability, control, and engineering-grade delivery.\n\u003c/p\u003e\n\n\u003cp\u003e\n  \u003cimg src=\"https://img.shields.io/badge/Workflow-Multi--Agent-111111?style=flat-square\" alt=\"Multi-Agent Workflow\" /\u003e\n  \u003cimg src=\"https://img.shields.io/badge/Architecture-Artifact--First-0A66C2?style=flat-square\" alt=\"Artifact First\" /\u003e\n  \u003cimg src=\"https://img.shields.io/badge/Validation-Gate--Guarded-8A2BE2?style=flat-square\" alt=\"Gate Guarded\" /\u003e\n  \u003cimg src=\"https://img.shields.io/badge/Agents-Claude%20Code%20%7C%20Gemini%20CLI%20%7C%20Codex%20CLI-2F855A?style=flat-square\" alt=\"Agents\" /\u003e\n  \u003cimg src=\"https://img.shields.io/badge/Python-Validator-3776AB?style=flat-square\u0026logo=python\u0026logoColor=white\" alt=\"Python Validator\" /\u003e\n\u003c/p\u003e\n\n\u003cp\u003e\n  Turn AI-assisted development from scattered chat into a durable operating system for research, planning, implementation, and verification.\n\u003c/p\u003e\n\n**[繁體中文](README.zh-TW.md)** | English\n\n\u003c/div\u003e\n\n---\n\n## Product Positioning\n\nConsilium Fabri is a multi-agent AI workflow framework designed to live inside the repository itself. It is not built around \"asking a model to code faster\"; it is built around creating a delivery system with explicit boundaries, reviewable checkpoints, durable artifacts, and hard verification.\n\nIt is especially useful when you need to:\n\n- keep engineering discipline while collaborating with AI\n- separate research, planning, implementation, and verification into explicit stages\n- prevent key decisions from disappearing into chat history\n- reduce the risk of untraceable, unreviewable, or unreproducible AI output\n- add an AI workflow layer to an existing project without adopting an entirely new platform\n\nThis project is not a prompt pack, and it is not a single-agent chat script. It is a workflow harness oriented toward engineering governance.\n\n---\n\n## Why This Project Exists\n\nMost multi-agent AI development breaks down in familiar ways:\n\n- research findings never land in a stable place\n- plans and implementation drift apart until ownership becomes unclear\n- verification stops at verbal claims instead of evidence\n- agent roles overlap and task boundaries become blurry\n- too much documentation gets stuffed into every prompt, increasing cost and instability\n\nConsilium Fabri exists to compress those failure modes into an explicit operating model with state, artifacts, and gates.\n\n---\n\n## Core Capabilities\n\n\u003ctable\u003e\n  \u003ctr\u003e\n    \u003ctd width=\"33%\" valign=\"top\"\u003e\n      \u003ch3\u003eMulti-Agent Collaboration\u003c/h3\u003e\n      \u003cp\u003eClaude Code, Gemini CLI, and Codex CLI each own a distinct responsibility so research, orchestration, and implementation stay focused instead of collapsing into a single blurry prompt.\u003c/p\u003e\n    \u003c/td\u003e\n    \u003ctd width=\"33%\" valign=\"top\"\u003e\n      \u003ch3\u003eArtifact First\u003c/h3\u003e\n      \u003cp\u003eEvery task is anchored in task, research, plan, code, verify, decision, and status artifacts rather than hidden chat memory, making the workflow traceable, reviewable, and restartable.\u003c/p\u003e\n    \u003c/td\u003e\n    \u003ctd width=\"33%\" valign=\"top\"\u003e\n      \u003ch3\u003eGate Validation\u003c/h3\u003e\n      \u003cp\u003eWorkflow gates and the validator enforce legal state transitions, required artifacts, and verification expectations so work cannot move forward on confidence alone.\u003c/p\u003e\n    \u003c/td\u003e\n  \u003c/tr\u003e\n\u003c/table\u003e\n\n---\n\n## Product Highlights\n\n### 1. Role Separation For Real Development Work\n- Claude Code acts as the orchestrator and workflow driver\n- Gemini CLI handles research and evidence gathering\n- Codex CLI handles implementation and delivery\n- Clear ownership reduces collisions, duplicated effort, and role drift\n\n### 2. A Strict Gate-Guarded Workflow\n- Tasks move through Intake, Research, Planning, Coding, Verification, and Done\n- Each stage has explicit prerequisites\n- Required steps cannot be skipped arbitrarily\n- Delivery becomes easier to review, replay, and audit\n\n### 3. Artifact-First Design You Can Audit\n- research findings live in research artifacts instead of chat summaries\n- implementation requires an approved plan artifact\n- verification requires a verify artifact\n- decisions can be recorded as decision artifacts\n- status is tracked in machine-readable files that support automation\n\n### 4. Validation As A Mechanism, Not A Slogan\n- `guard_status_validator.py` is built in\n- `guard_contract_validator.py` is built in\n- legal state transitions can be checked automatically\n- required artifacts, metadata, and research / PDCA contracts can be checked automatically\n- root / `template/` / Obsidian workflow drift can be checked automatically\n- it reduces the risk of work being declared done without being genuinely verified\n\n### 5. A More Disciplined Context Loading Strategy\n- agents do not need to read the entire documentation set on every run\n- documentation is loaded by role and phase\n- token usage stays lower and more predictable\n- prompt pollution and instability are reduced across longer task chains\n\n### 6. Documentation And Timestamp Discipline\n- long-lived Markdown defaults to Traditional Chinese (Taiwan) unless a specific exception is needed\n- commands, file paths, placeholders, schema literals, and status values remain in English\n- recorded times and `Last Updated` values must use `Asia/Taipei` in ISO 8601 format with `+08:00`\n- root docs, `template/` docs, and Obsidian entry docs must stay semantically aligned\n\n### 7. Clear Guard Boundaries\n- `guard_status_validator.py` validates task / artifact / state rules\n- plan/code scope drift is now a default hard failure: dirty task-owned files are checked against actual git changed files, clean tasks can replay pinned `commit-range` evidence, use an `archive fallback` via `Archive Path` / `Archive SHA256` when git objects are gone, or use `github-pr` evidence against the GitHub PR files API; `Snapshot SHA256` still guards the reconstructed file list, `GITHUB_TOKEN` / `GH_TOKEN` covers private or rate-limited GitHub access, and `--allow-scope-drift` still only downgrades true drift, not corrupted evidence\n- `guard_contract_validator.py` validates workflow docs, bootstrap rules, template sync, and Obsidian sync\n- when `CLAUDE.md` / `GEMINI.md` / `CODEX.md` changes, prompt regression cases must be updated together\n- a workflow rule change is incomplete until README, `template/`, and Obsidian entry docs are updated together\n\n### 8. Built-In Red-Team Exercises\n- `docs/red_team_runbook.md` defines the static attacks, live drills, and replay workflow\n- `docs/red_team_scorecard.md` provides the scoring matrix\n- `docs/red_team_backlog.md` tracks follow-up hardening work\n- `python artifacts/scripts/run_red_team_suite.py --phase all` reruns the built-in red-team suite and live drill samples\n- `python artifacts/scripts/prompt_regression_validator.py --root .` runs fixed prompt regression cases for `CLAUDE.md`, `GEMINI.md`, `CODEX.md`, and critical workflow contracts\n- the fixed prompt regression suite now also covers artifact-only truth/completion, workflow sync completeness, Gemini blocked preconditions, Codex summary discipline, conflict-to-decision routing, decision schema integrity, external failure STOP, decision-gated scope waivers, historical diff evidence contracts, pinned diff evidence integrity, GitHub provider-backed diff evidence, and archive retention fallback contracts\n- `python artifacts/scripts/run_red_team_suite.py --phase prompt` runs prompt regression through the same report pipeline\n\n---\n\n## Use Cases\n\nThis project is especially suitable for:\n\n| Use Case | Description |\n|---|---|\n| Personal AI development framework | A solo developer can still manage AI collaboration with engineering discipline |\n| Small team collaboration | Build a controlled workflow without adopting a large platform |\n| Traceable AI delivery | Preserve a full trail across research, planning, implementation, and verification |\n| Existing repository adoption | Add this as a workflow layer to an existing repo |\n| Open source showcase | Demonstrate a practical methodology for AI-assisted engineering |\n\n---\n\n## Workflow Overview\n\n```text\nIntake\n  |\n  v\nResearch\n  |\n  v\nPlanning\n  |\n  v\nCoding\n  |\n  v\nVerification\n  |\n  v\nDone\n```\n\nThe model is simple on purpose: each stage produces the artifact that justifies the next stage. That keeps collaboration inspectable and prevents \"magic progress\" that only exists inside a chat transcript.\n\n---\n\n## Getting Started\n\n### Prerequisites\n\n- **Python 3.10+** (for validator scripts)\n- **Git** (version control)\n- **Claude Code** (orchestrator agent — via VS Code extension or CLI)\n- **Gemini CLI** (research agent — optional, for full workflow)\n- **Codex CLI** (implementation agent — optional, for full workflow)\n- **PyYAML** (`pip install -r requirements.txt`)\n\n### Quick Start — New Project\n\n```bash\n# 1. Clone the template into your project\ngit clone https://github.com/arcobaleno64/consilium-fabri.git my-project\ncd my-project\n\n# 2. Replace placeholders in CLAUDE.md (remove fork section if not needed)\n#    {{PROJECT_NAME}}, {{REPO_NAME}}, {{UPSTREAM_ORG}}\n\n# 3. Bootstrap validation\npython artifacts/scripts/guard_status_validator.py --task-id TASK-900 --auto-classify\npython artifacts/scripts/update_repository_profile.py\npython artifacts/scripts/guard_contract_validator.py --check-readme\npython artifacts/scripts/guard_contract_validator.py\npython artifacts/scripts/prompt_regression_validator.py --root .\n\n# 4. (Optional) Run the red-team suite\npython artifacts/scripts/run_red_team_suite.py --phase all\n```\n\nSee `BOOTSTRAP_PROMPT.md` for the full bootstrapping guide.\n\n### Quick Start — Existing Project\n\nCopy the `template/` directory contents into your repository root, replace placeholders, and run the same bootstrap validation commands above.\n\n---\n\n## Repository Structure\n\n```\n.\n├── AGENTS.md                  # Document index and phase-loading matrix\n├── CLAUDE.md                  # Orchestrator (Claude Code) entry file\n├── GEMINI.md                  # Research agent (Gemini CLI) entry file\n├── CODEX.md                   # Implementation agent (Codex CLI) entry file\n├── OBSIDIAN.md                # Obsidian vault entry note\n├── BOOTSTRAP_PROMPT.md        # New project bootstrapping guide\n├── README.md / README.zh-TW.md\n├── requirements.txt           # Python dependencies (PyYAML)\n│\n├── docs/                      # Workflow specification documents\n│   ├── orchestration.md       # Full workflow: goals, principles, stages, gates\n│   ├── artifact_schema.md     # 8 artifact type schemas (§5.1–§5.8)\n│   ├── workflow_state_machine.md  # 8 states + legal transitions\n│   ├── premortem_rules.md     # Risk analysis format + quality guardrails\n│   ├── subagent_roles.md      # 7 agent role definitions\n│   ├── subagent_task_templates.md\n│   ├── lightweight_mode_rules.md\n│   ├── red_team_runbook.md    # Red-team exercise playbook\n│   ├── red_team_scorecard.md  # Scoring matrix\n│   ├── red_team_backlog.md    # Hardening backlog\n│   └── templates/             # Subagent task prompt templates\n│\n├── artifacts/                 # All workflow artifacts (the single source of truth)\n│   ├── tasks/                 # Task artifacts\n│   ├── research/              # Research artifacts\n│   ├── plans/                 # Plan artifacts\n│   ├── code/                  # Code artifacts\n│   ├── verify/                # Verification artifacts\n│   ├── decisions/             # Decision artifacts\n│   ├── improvement/           # Improvement artifacts\n│   ├── status/                # Machine-readable status + decision registry\n│   ├── red_team/              # Red-team exercise reports\n│   └── scripts/               # Validator and automation scripts\n│       ├── guard_status_validator.py\n│       ├── guard_contract_validator.py\n│       ├── prompt_regression_validator.py\n│       ├── run_red_team_suite.py\n│       ├── repo_health_dashboard.py\n│       ├── build_decision_registry.py\n│       └── drills/            # Prompt regression test cases\n│\n├── .github/\n│   ├── copilot-instructions.md    # VS Code Copilot global rules\n│   ├── repository-profile.json   # GitHub About / Topics profile\n│   ├── memory-bank/               # Stable reference knowledge base\n│   ├── prompts/                   # Prompt and skill files\n│   ├── agents/                    # Agent definition files\n│   ├── skills/                    # Skill metadata\n│   └── workflows/                 # GitHub Actions CI\n│\n├── template/                  # Clean template for new projects (sync target)\n└── external/                  # External project integrations\n```\n\n---\n\n## Validator Commands\n\n| Command | Purpose |\n|---|---|\n| `python artifacts/scripts/guard_status_validator.py --task-id TASK-XXX` | Validate task state, artifacts, and scope drift |\n| `python artifacts/scripts/guard_status_validator.py --task-id TASK-XXX --auto-classify` | Auto-classify task as lightweight or full-gate |\n| `python artifacts/scripts/guard_contract_validator.py` | Validate root ↔ template ↔ Obsidian sync |\n| `python artifacts/scripts/guard_contract_validator.py --check-readme` | Validate README structure compliance |\n| `python artifacts/scripts/prompt_regression_validator.py --root .` | Run prompt regression test cases |\n| `python artifacts/scripts/run_red_team_suite.py --phase all` | Run the full red-team exercise suite |\n| `python artifacts/scripts/run_red_team_suite.py --phase prompt` | Run prompt regression via the report pipeline |\n| `python artifacts/scripts/repo_health_dashboard.py` | Generate repository health dashboard |\n| `python artifacts/scripts/build_decision_registry.py --root .` | Rebuild the decision registry |\n| `python artifacts/scripts/update_repository_profile.py` | Update GitHub repository profile |\n\n---\n\n## Context System\n\nThis project includes a layered context management system for VS Code Copilot:\n\n- **`.github/copilot-instructions.md`** — Global stable rules, auto-loaded by VS Code\n- **`.github/memory-bank/`** — Stable reference knowledge (artifact rules, workflow gates, prompt patterns, project facts)\n- **`.github/prompts/`** — Task-scoped prompts (pack-context, context-review, remember-capture)\n- **`.github/skills/`** — Reusable skill definitions (always-ask-next)\n\nAgents load documentation by role and phase, not all at once. See `AGENTS.md` for the phase-loading matrix.\n\n---\n\n## Contributing\n\n1. Fork the repository\n2. Create a feature branch\n3. Follow the artifact-first workflow: task → research → plan → code → verify\n4. Run validators before submitting:\n   ```bash\n   python artifacts/scripts/guard_contract_validator.py\n   python artifacts/scripts/prompt_regression_validator.py --root .\n   ```\n5. Open a Pull Request\n\nAll workflow documentation defaults to Traditional Chinese (Taiwan). Commands, file paths, placeholders, schema literals, and status values remain in English.\n\n---\n\n## License\n\nThis project is licensed under the [MIT License](LICENSE).\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Farcobaleno64%2Fconsilium-fabri","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Farcobaleno64%2Fconsilium-fabri","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Farcobaleno64%2Fconsilium-fabri/lists"}