# EPURE (Giskard-AI/epure)

> Engineered Planning, Unified Review & Execution (FR: Élaboration de Plans, Usinage, Revue et Exécution) — Open-source GitHub-first specification-driven software engineering framework built on Reusable Workflows, TypeScript Custom Actions (Node.js 24 LTS), and AI Agent Skills.

- Documentation Home (EN): https://giskard-ai.github.io/epure/
- Documentation Accueil (FR): https://giskard-ai.github.io/epure/fr/
- Full Concatenated Corpus for LLMs: https://giskard-ai.github.io/epure/llms-full.txt
- GitHub Repository: https://github.com/Giskard-AI/epure
- License: Apache-2.0

## Core Summary for AI Agents

- **Preferred Onboarding (`/setup-repo`)**: Invoke `/setup-repo <owner>/<repo>` in any Skills-compatible coding agent (`.agents/skills/setup-repo/SKILL.md`). The agent inspects the repository stack and domains, runs `scripts/wizards/setup-github-apps-wizard.sh --repo <owner>/<repo> --llm-env <env-name>` (default `epure-agent`), executes `node dist/cli/setup-repo.mjs --dry-run` then `--apply --apply-github-settings`, configures 3 Repository Rulesets (`epure-feature-governance`, `epure-dev-governance`, `epure-release-immutability`), generates `.github/zizmor.yml` and a hardened `renovate.json` (with 3-to-14-day quarantine cooldowns), and dynamically composes specialist reviewers in `reviewers_dir` (default `.agents/reviewers/*.md`).
- **Four Mandatory Human Gates**:
  1. Initiating `/epure plan` and answering the architectural Grill on the GitHub Issue.
  2. Approving the locked specification mandate via `/epure approve` (integrating `plan/*` directly into `feature/*` and locking `spec_sha`).
  3. Reviewing and squash-merging the final `feature/<issue>-<slug> -> dev` Pull Request.
  4. Triggering production SemVer releases (`dev -> main` fast-forward) via `release-promote.yml` / `release.yml`.
- **Security & Execution Invariants (33 ADRs & 12 Bilingual Archify Diagrams)**: Out-of-branch atomic CAS state (`refs/epure/state/<issue>`), Zizmor-compliant GitHub Environment secret scoping (`<llm-env>` and `release`), 2-phase Ubuntu 26.04 Linux kernel sandbox (`landrun` + `bubblewrap` with stateless `rtk` condensation), event-driven external CI continuation (`waiting_ticket_ci` / `waiting_feature_ci`), transactional incremental file tools (`StagedTicketImplementationBuffer`), zero-context `EpicConvergenceAudit`, Reviewer Roster always loaded from trusted `dev` (never from untrusted PR head), Dual GitHub Apps (`Agent App` vs `Release App`), and `iFixAi` trajectory auditing.

## English Documentation

- [Introduction & Naming (L'Épure)](https://giskard-ai.github.io/epure/introduction): Philosophy of L'Épure, 4 mandatory human validation gates, 3-category step taxonomy, and upstream open-source attributions.
- [Installation & /setup-repo Skill](https://giskard-ai.github.io/epure/installation): Primary onboarding via the /setup-repo agent skill, Dual GitHub Apps wizard with environment-scoped secrets (--llm-env and release), 3 Repository Rulesets, and dynamic reviewer roster design (reviewers_dir, default .agents/reviewers/).
- [Basic Usage (Case Study: craft-calculator)](https://giskard-ai.github.io/epure/basic-usage): End-to-end walkthrough of an issue from /epure plan Grill interview and Goldfish spec audit to native GitHub sub_issues/blocked_by DAG execution, 2-phase Ubuntu 26.04 kernel sandbox, event-driven CI continuation, EpicConvergenceAudit, and human merge.
- [Commands Reference (/epure & /setup-repo)](https://giskard-ai.github.io/epure/commands): Complete reference for /setup-repo, /epure plan, /epure to-spec, /epure to-tickets, /epure approve, /epure continue, /epure sync, /epure status, /epure cancel, and the epure-ci-actions-flow pipeline.
- [Configuration (.github/agents.yml & renovate.json)](https://giskard-ai.github.io/epure/configuration): Flat canonical declarative configuration reference for .github/agents.yml (models, providers, verification.targets, web_tools, security_invariants), configurable <llm-env> GitHub Environment, and supply-chain security policy in renovate.json & .github/zizmor.yml.
- [Architecture, Security & Multi-Agent Review](https://giskard-ai.github.io/epure/architecture): Dual-App privilege separation, out-of-branch CAS state in refs/epure/state/<issue>, 3-stage verification pipeline (Ubuntu 26.04 bwrap/landrun kernel sandbox & stateless rtk), event-driven CI continuation, incremental workspace mutation tools, and EpicConvergenceAudit.
- [How-To: Onboard a Repository](https://giskard-ai.github.io/epure/guides/how-to-onboard-repo): Step-by-step guide to bootstrapping a consumer repository with /setup-repo and environment-scoped secrets.
- [How-To: Drive an Issue End-to-End](https://giskard-ai.github.io/epure/guides/how-to-drive-an-issue): Practical guide to running the Grill, approving the spec_sha mandate, event-driven CI waiting, /epure continue, /epure sync, Phase 2 convergence tickets, and merging feature/* -> dev.
- [How-To: Customize Reviewers & Impeccable Gate](https://giskard-ai.github.io/epure/guides/how-to-customize-reviewers): How to add domain-specific reviewers in the reviewer roster directory (reviewers_dir, default .agents/reviewers/*.md) and configure DESIGN.md + impeccable-gate.
- [How-To: Promote a Release (dev -> main)](https://giskard-ai.github.io/epure/guides/how-to-release): How to promote dev to main via fast-forward under the Release App and handle hotfix back-merges.
- [Domain Glossary (EN / FR)](https://giskard-ai.github.io/epure/reference/glossary): Canonical Ubiquitous Language dictionary for EPURE (Q1–Q111, ADR-0029–ADR-0033, v1.7.0).

## Documentation Française

- [Introduction & Nom (L'Épure)](https://giskard-ai.github.io/epure/fr/introduction): Philosophie de L'Épure, 4 portes de validation humaine obligatoires, taxonomie à 3 catégories et crédits open-source.
- [Installation & Skill /setup-repo](https://giskard-ai.github.io/epure/fr/installation): Onboarding recommandé via le skill /setup-repo, wizard des 2 GitHub Apps avec secrets cloisonnés par environnement (--llm-env et release), 3 Rulesets et composition dynamique du roster de reviewers (reviewers_dir, .agents/reviewers/ par défaut).
- [Usage de Base (Étude de Cas : craft-calculator)](https://giskard-ai.github.io/epure/fr/basic-usage): Parcours complet d'une issue depuis l'entretien Grill /epure plan et l'audit Goldfish jusqu'au DAG natif sub_issues/blocked_by, au sandbox noyau Ubuntu 26.04, à la CI événementielle, à l'EpicConvergenceAudit et au merge humain.
- [Référence des Commandes (/epure & /setup-repo)](https://giskard-ai.github.io/epure/fr/commands): Référence complète du skill /setup-repo, des commandes /epure (/epure plan, to-spec, to-tickets, approve, continue, sync, status, cancel) et du pipeline epure-ci-actions-flow.
- [Configuration (.github/agents.yml & renovate.json)](https://giskard-ai.github.io/epure/fr/configuration): Référence déclarative du schéma plat .github/agents.yml, environnement GitHub <llm-env> configurable et politique supply-chain Renovate & Zizmor.
- [Architecture, Sécurité & Revue Multi-Agents](https://giskard-ai.github.io/epure/fr/architecture): Séparation des privilèges à 2 GitHub Apps, état CAS hors-branches, pipeline en 3 étapes (sandbox noyau Ubuntu 26.04 bwrap/landrun & rtk sans état), continuation CI événementielle, outils de mutation incrémentale et EpicConvergenceAudit.
- [Guide : Onboarder un Dépôt](https://giskard-ai.github.io/epure/fr/guides/how-to-onboard-repo): Guide pas à pas pour équiper un dépôt consommateur avec /setup-repo et des secrets cloisonnés par environnement.
- [Guide : Piloter une Issue de Bout en Bout](https://giskard-ai.github.io/epure/fr/guides/how-to-drive-an-issue): Conduire le Grill, approuver le mandat spec_sha, gérer la CI événementielle, /epure continue, /epure sync, les tickets de Phase 2 et fusionner feature/* -> dev.
- [Guide : Personnaliser le Roster & Impeccable](https://giskard-ai.github.io/epure/fr/guides/how-to-customize-reviewers): Créer des reviewers spécialisés dans le répertoire du roster (reviewers_dir, .agents/reviewers/*.md par défaut) et configurer DESIGN.md.
- [Guide : Promouvoir une Release (dev -> main)](https://giskard-ai.github.io/epure/fr/guides/how-to-release): Promotion fast-forward de dev vers main via la Release App.