Metadata-Version: 2.4
Name: hrevn-anthropic-runner
Version: 0.1.0a0
Summary: HREVN CLI runner for Claude Code and Anthropic-facing workflows.
Author-email: Miguel Herrero <contact@hrevn.com>
License-Expression: MIT
Project-URL: Homepage, https://hrevn.com
Project-URL: Repository, https://github.com/ai-human-andalusia/hrevn-surface-anthropic
Keywords: hrevn,anthropic,claude-code,workflow-integrity,auditability
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3 :: Only
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Requires-Python: >=3.11
Description-Content-Type: text/markdown
License-File: LICENSE
Dynamic: license-file

# HREVN for Claude Code - Verifiable Workflow State via MCP and Skills

This repo is the Anthropic-facing HREVN surface for Claude Code: skills, a local
runner, and an MCP path for resumable workflows, tamper-evident receipts, and
audit-ready traceability against the live HREVN managed runtime.

## Why HREVN

AI agents and multi-step workflows fail in ambiguous ways. When a sequence is
interrupted, neither the user nor the system can always determine with
certainty what completed, what failed mid-execution, and where work can safely
resume. Without a verifiable record, context is reconstructed from memory or
chat history, wasting tokens, repeating work, and leaving no reliable trail.

HREVN adds a structured evidence layer: baseline checks before consequential
steps, tamper-evident receipts after execution, and manifests that allow
workflows to continue from the last verified point rather than restarting from
scratch.

For teams operating in regulated or high-stakes environments, HREVN also
supports evidentiary discipline: structured records of what ran, under what
authority, and when it stopped. This is particularly relevant for AI systems
that may fall within EU regulatory timelines in 2026 and beyond. HREVN does
not make a system legally compliant, but it provides structured, verifiable
evidence that compliance, audit, and governance processes can use.

In Anthropic-facing workflows, HREVN extends the skill layer with structured
records of what ran, under what authority, and where execution can safely
resume through the runner bridge to the managed runtime.

## What it is
- an Anthropic-facing skill repo
- a thin public surface
- a local runner bridge to `https://api.hrevn.com`
- an MCP-ready path for Claude Code tool use

## What it is not yet
- not a native remote-skill marketplace package
- not yet a polished end-user MCP distribution
- not the private HREVN runtime
- not the private commercial toolkit

## Skill chain
1. `hrevn-baseline-check`
2. `hrevn-sign-on-complete`
3. `hrevn-verify-bundle`

## Quick Start

```bash
pipx install hrevn-anthropic-runner
export HREVN_API_KEY="<issued-alpha-key>"
hrevn-anthropic health-check
hrevn-anthropic self-test
hrevn-anthropic baseline
```

If you prefer the repo checkout instead, you can still do:

```bash
git clone https://github.com/ai-human-andalusia/hrevn-surface-anthropic
cd hrevn-surface-anthropic
pipx install .
```

If `pipx` is not available, a local fallback is:

```bash
python3 -m pip install .
```

The installable command should return a real `BaselineResult` JSON once auth is
set correctly.

For the MCP-first and guided alpha path, see:
- `docs/ANTHROPIC_ALPHA_TESTING.md`
- `docs/MCP_USAGE.md`

Alpha keys are issued out-of-band for testing rather than embedded in the
public repo.

## Optional MCP path

If you want Claude Code to discover HREVN as MCP tools rather than only through
skill guidance plus a local runner, use the companion MCP server at:

- `https://github.com/ai-human-andalusia/hrevn-mcp-server`

See:
- `docs/MCP_USAGE.md`
- `docs/claude_code_mcp_config.example.json`

## Test in Claude Code

1. Open Claude Code in this repository.
2. Make sure `HREVN_API_KEY` is exported in the shell environment Claude Code will use.
3. First verify the runner directly:

```bash
hrevn-anthropic self-test
hrevn-anthropic baseline
```

4. Then ask Claude Code:

> Run a baseline check on `examples/baseline_check_request.json`

The expected path is:

```text
skill instructions -> local runner -> https://api.hrevn.com
```

## What is already proven in this repo
- the local runner can call the live managed API
- the baseline-check example returns a real `BaselineResult`
- the current path is usable today without waiting for MCP
- the Anthropic surface now has a documented MCP upgrade path for the same backend

## Included
- `.claude/skills/...`
- `docs/MANAGED_API_USAGE.md`
- `pyproject.toml`
- `scripts/hrevn_anthropic_runner.py`
- `src/hrevn_anthropic_runner/...`
- `examples/baseline_check_request.json`

## Managed Runtime Bridge
The live managed endpoint is:
- `https://api.hrevn.com`

Current path:
- skill guidance now
- local runner bridge now
- MCP available as the next bridge when Claude Code is configured for it

## Current status
This is a public Anthropic surface candidate with a real testing path today.
It is intentionally honest about the current runner-first path while also exposing
an MCP route for Anthropic-first testing.

## Rule
The skills guide behavior, but the runtime truth stays in the managed API and private HREVN core.
The skill itself does not natively perform remote HTTPS on its own.
