Workflow principles
These principles govern A2C execution in every A2C-based repository.
Docs first
Work begins with documentation and architecture rather than code generation. Typical early artifacts:
- README and docs structure
- Architecture overview
- ADR README and index
- Templates, rules, and prompts
- Commit plan
Architecture as contract
Architecture docs, specs, and accepted ADRs are governing artifacts. AI must use them as the contract — not as loose inspiration.
When a spec and implementation disagree, stop and reconcile before proceeding.
Small bounded steps
Implementation happens in small, cohesive, reviewable, commit-sized slices. A2C explicitly rejects big-bang generation.
Each batch has a defined scope and a stop condition. Do not continue into the next batch without review.
Repository-native workflow memory
The repository stores rules, templates, prompts, plans, profiles, docs structure, and workflow metadata. This makes the process:
- reusable across projects
- consistent for AI across sessions
- inspectable in Git history
AI as force multiplier
AI performs scaffolding, documentation plumbing, repetitive configuration, refactoring assistance, and CI wiring inside explicit constraints.
The developer keeps control of architecture, quality bar, and review.
For the full mental model — architecture that works without AI, contracts before models, designer vs user roles — see ai-principles.md.
ADR-first architectural history
Significant decisions are captured in ADRs and committed with the dedicated adr(<category>): ... type. One ADR per commit. See ../adr/README.md.
When ASRs are used, treat them as first-class contracts with the same change discipline; see asr-adr-pr-governance.md.
Traceability
Commit plans connect batches to architecture and ADRs. Git history should tell a clear story: docs → decisions → scaffolding → implementation.
Changelog (CHANGELOG.md) provides curated release-level history — what changed in each version — separate from commit logs and ADR rationale. See changelog-policy.md.
Related
- ai-principles.md — short AI mental model (human overview)
- Encoded in pathname:///a2c-assets/rules/ for machine consumption.