K Tool · Open Starter Kit

# K Components Starter

K Components Starter is a minimal open-source package for turning agent work
into governed, auditable, evidence-bound component workflows.

The package is intentionally small. It opens the public starter layer first:
component contracts, a tiny offline optimizer, example tasks, validation
commands, and launch material for GitHub and Hugging Face. It does not publish
the full internal K research system or private operational ledgers.

## What This Helps With

- Add a component protocol around agent tools, gates, workflows, evaluators,
  and audit ledgers.
- Route sensitive agent tasks to review before execution.
- Convert a plain agent task into a K-governed plan with validation evidence.
- Produce a governance scorecard without claiming model intelligence gains.
- Give users a reproducible first package they can run locally in minutes.

## Quick Start

```bash
python -B tools/validate_package.py --write
set PYTHONPATH=src
python -B -m unittest discover -s tests -p "test_*.py"
python -B -m kcomponents.cli metrics --root .
python -B -m kcomponents.cli optimize examples/basic_agent_task.json --root .
```

On PowerShell, the one-command validation path is:

```powershell
powershell -NoProfile -ExecutionPolicy Bypass -File 99_VALIDATE_OPEN_SOURCE_STARTER.ps1
```

## Starter Components

| Component | Type | Purpose |
| --- | --- | --- |
| `k.policy.component_protocol_v2` | K-Policy | Public starter contract for governed K components. |
| `k.gate.agent_safety_gate` | K-Gate | Routes sensitive or high-risk agent tasks to review. |
| `k.workflow.agent_task_optimizer` | K-Workflow | Converts a task into a gated and auditable plan. |
| `k.evaluator.agent_run_scorecard` | K-Evaluator | Scores traceability, gate coverage, evidence binding, and claim control. |
| `k.audit.trace_ledger` | K-Audit | Gives agent decisions and outputs a hashable trace surface. |

## Claim Boundary

This package currently claims a governance and reproducibility improvement:
clearer component boundaries, stronger review routing, better traceability, and
explicit evidence requirements. It does not claim higher model intelligence,
leaderboard superiority, production safety certification, or live external
agent benchmark results.

## Public Release Strategy

The open-source path is staged:

- V0.1: starter protocol, offline optimizer, validation tool, launch docs.
- V0.2: adapters for common agent frameworks.
- V0.3: public benchmark harness and trace examples.
- V0.4: memory and RAG governance components.
- V0.5: experimental tensor-memory sidecar interface.
- V1.0: stable protocol with third-party examples and evidence-backed case
  studies.

See `docs/OPEN_SOURCE_ROADMAP.md` for the full phased plan.

## Repository Targets

- GitHub: engineering home, issues, examples, CI, releases, discussions.
- Hugging Face: discoverability home for a Space, Dataset, or paper-style
  project card that explains how K components wrap agent workflows.

See `docs/GITHUB_LAUNCH_PLAN.md` and `docs/HUGGINGFACE_LAUNCH_CARD.md`.

## License

Code is released under MIT. Documentation is released under CC BY 4.0 unless a
file states otherwise.

This English reading preview is for reference. Downloads retain their original content and licenses.