Agent Bootstrap Preview

Agent Bootstrap Preview is a public, sanitized snapshot of a local-first governance harness for coding agents.

It is meant for review and evaluation. It shows how a coding agent can be anchored to GitHub issue authority, workspace-scoped task context, validation evidence, and finish gates before it edits durable repository files.

Current preview release: v0.3.0-preview.1

Source anchor: ClarusIubar/agent_bootstrap@v1.4.0-rc.1

Runtime anchor: ClarusIubar/agent-governance-runtime@v0.1.1

Runtime access note: the runtime repository and wheel asset are private. This preview includes the consumer surface and documentation boundary, but users without private access cannot download the runtime wheel directly from GitHub.

What This Preview Is

• A public snapshot of the governance CLI, policies, skills, shim templates, and reviewer-safe tests.
• A reference for local governance compliance: issue-bound hydration, edit gates, qmd lookup boundaries, runtime doctor/install surfaces, and finish readback.
• A sanitized distribution view. It does not preserve the operational source repository history or private completion evidence.

What This Preview Is Not

• It is not the operational private repository.
• It is not a complete public runtime distribution while the runtime wheel stays private.
• It is not a multi-machine control plane, cross-device lease system, Hermes learning loop, MCP/App server, daemon, or autonomous multi-agent orchestrator.
• It does not replace GitHub issues, PRs, CI, or merge readback as source of truth.

Quick Start For Reviewers

git clone https://github.com/ClarusIubar/agent_bootstrap-preview.git
cd agent_bootstrap-preview
python -m pip install -r requirements-dev.txt
python -m pytest tests -q
python scripts/agent_bootstrap.py --help

On Unix-like shells:

git clone https://github.com/ClarusIubar/agent_bootstrap-preview.git
cd agent_bootstrap-preview
python -m pip install -r requirements-dev.txt
python -m pytest tests -q
python scripts/agent_bootstrap.py --help

How To Point An Agent At A Governed Repo

If this harness is installed in a private governance home, give an agent a short instruction like this:

This repo uses a local governance harness. Before editing, run the governance
doctor/install path, then hydrate the current workspace with context rehydrate
or context start. Do not implement unless the output says task-hydrated and
Edit gate: implementation-ready.

For an operational private clone, the usual entrypoints are:

scripts\agent-governance.cmd runtime doctor
scripts\agent-governance.cmd qmd doctor --workspace <workspace>
scripts\agent-governance.cmd context rehydrate --skip-sync --workspace <workspace>
scripts\agent-governance.cmd context start --workspace <workspace> --issue <child-issue-url-or-number>
scripts\agent-governance.cmd context finish --workspace <workspace> --issue <child-issue-url-or-number> --dry-run

These commands are shown for the governance workflow. Runtime wheel installation may require access to the private runtime release asset.

Core Workflow

1. A GitHub parent issue defines the durable roadmap.
2. A child issue defines executable scope, out-of-scope items, validation commands, and architecture boundary evidence.
3. context start hydrates that child issue into the local workspace.
4. The edit gate opens only when repository, branch, child issue, checklist, policy, architecture boundary, and validation contract agree.
5. The agent changes only files inside the hydrated scope.
6. Local validation, CI, and readback checks run before completion.
7. context finish --dry-run proves the completion plan.
8. context finish --apply records finalization and local completion only after GitHub readback succeeds.

Authority Boundary

• GitHub issue/checklist/PR/CI/merge readback is source of truth.
• Local registry/focus, qmd, Hermes, Graphify, runtime projections, and Wiki pages are cache, context, evidence, or readback aids.
• qmd/Hermes/Graphify/runtime availability never grants edit authority by itself.
• Public preview documentation is review material, not completion authority.

Included Surfaces

• scripts/: governance CLI and helper modules.
• policies/: reviewer-facing policy documents used by generated shims.
• registry/skills/: reusable governance-aware skills and shared pointers.
• templates/shims/: managed shim templates.
• tests/: public validation and regression tests included in the snapshot.
• docs/GOVERNANCE_INDEX.md and docs/PUBLIC_PREVIEW_SCOPE.md: public routing and sanitization notes.

Excluded Surfaces

• Operational Git history and fork linkage.
• Private issue, PR, CI, merge, and completion evidence ledgers.
• Machine-local state under .local/.
• .tmp-wiki, caches, coverage outputs, and temporary files.
• Token-like fixtures and non-public runner traces.
• Private runtime wheel access credentials.

Validation For v0.3.0-preview.1

• Source anchor: ClarusIubar/agent_bootstrap@v1.4.0-rc.1
• Runtime anchor: ClarusIubar/agent-governance-runtime@v0.1.1
• Bootstrap validation: 3539 passed, 3 skipped, 381 subtests passed
• Bootstrap coverage: total 98.46%, branch 97.45%
• Runtime validation: 455 passed
• Public preview smoke tests: 6 passed
• Runtime wheel: agent_governance_runtime-0.1.1-py3-none-any.whl
• Runtime wheel SHA-256: 05e88dda504d8f3dbf682dd459a25f7809b595292135207f92156ea06d5ad37f

Wiki

• Preview Wiki
• Preview Scope
• Reviewer Quick Start
• Architecture Boundary
• Workflow Overview

License

This project is distributed under the MIT License. See LICENSE.