harness-template is currently in harness-bootstrap mode. The repository is being prepared for agent-first engineering before product code is added.
The immediate goal is to make the codebase legible, enforceable, and easy for coding agents to operate in:
- A short root
AGENTS.mdpoints to repository-local sources of truth docs/design-docs/harness-template.mdexplains the repository's harness template and operating modelscripts/bootstrap_codebase.pycan generate repo-owned starter code for supported stacksdocs/holds versioned product, design, reliability, and planning knowledgedocs/exec-plans/is the home for active and completed execution plansscripts/check_harness_readiness.pymechanically validates the scaffold.github/workflows/harness-readiness.ymlruns the check in CI
- Read
AGENTS.md - Read
ARCHITECTURE.md - Read
docs/design-docs/harness-template.md - Read the relevant docs in
docs/ - Create an execution plan for any non-trivial task
- Run
python3 scripts/check_harness_readiness.pywhen changing scaffold docs - Use
python3 scripts/bootstrap_codebase.py --helpwhen adding a supported starter template
.
├── AGENTS.md
├── ARCHITECTURE.md
├── README.md
├── templates/
├── docs/
├── scripts/
└── .github/workflows/
The next repository milestone should add:
- Product specs for the first user-facing workflows
- Additional repo-owned stack templates beyond
web-nextjs - Architectural invariants that can be enforced mechanically
- Agent-operable local dev tooling for validation and observability
The harness template follows an agent-first approach:
- Keep root instructions short and move durable policy into repository-local docs
- Treat execution plans as persistent working memory for non-trivial changes
- Prefer monorepo-ready structure so future stacks can be added without churn
- Add mechanical checks early so repository discipline does not depend on memory
- Grow policy in small, reviewable increments instead of writing speculative process docs