# rein-aharness

(formerly `agent-harness`, renamed per glas-harness ADR-001 "rein" harness family)

The single shared runtime for unattended agent instances in the Coulomb /
Operational Knowledge ecosystem. Projects declare agents (manifest +
`.kaizen/` state); this harness runs them — blueprint via kaizen-agentic,
tasks via activity-core, compute via llm-connect, credentials via
OpenBao/ops-warden, reporting to the Custodian State Hub.

Within the glas-harness meta-framework, this repo is the **`rein-aharness`
rein**: the Claude-Code-CLI-driven, unattended/scheduled harness for
tenants that need governed cron-style agent runs. See
[glas-harness ADR-001](../glas-harness/docs/adr/ADR-001-rein-harness-family.md)
for how it relates to other reins (e.g. `rein-openweights`).

The CLI command (`rein-aharness`) and Python package (`rein_aharness`) are
renamed. Deploy artifacts (Docker image tag, k8s namespace, Railiance host
paths) are renamed in this repo too, but the live Railiance cutover itself
(moving the host-side checkout + secrets dir, redeploying) is a separate,
deliberate step — tracked in
[workplans/HARNESS-WP-0002](workplans/HARNESS-WP-0002-rename-and-glas-harness-alignment.md)
and [deploy/README.md](deploy/README.md)'s rename cutover checklist.

- Why and boundaries: [INTENT.md](INTENT.md)
- Decision + architecture: [docs/adr/ADR-001](docs/adr/ADR-001-agent-harness-architecture.md)
  (kept under its original filename — it documents the pre-rename decision),
  [docs/architecture.md](docs/architecture.md)
- Current work: [workplans/](workplans/)

## Usage

```
# Validate a consuming repo's instance manifest (.kaizen/schedule.yml)
rein-aharness validate --target ~/binky-control
rein-aharness validate --target ~/binky-control --strict   # require harness fields

# List named tool profiles
rein-aharness profiles

# Run exactly one task (local JSON task-file path — dev)
rein-aharness run --task-file examples/task-hello-sandbox.json [--no-hub] [--no-metrics]

# Execution-profile overrides supplied by glas-harness
rein-aharness run --task-file examples/task-hello-sandbox.json \
  --model claude-sonnet-4-6 \
  --tool-profile green-commit-only \
  --budget-tokens 60000 \
  --no-hub --no-metrics

# Primary production intake: activity-core ops_run claim (ACT-ADR-005 / REIN-A-0002)
export ACTIVITY_CORE_URL=http://127.0.0.1:8010
export ACTIVITY_CORE_WORKER_TOKEN=…
export AGENT_HARNESS_REPO_MAP='{"freedom-intelligence":"~/freedom-intelligence","binky-control":"~/binky-control"}'
rein-aharness poll --source=ops-run --no-claim
rein-aharness run --from-ops-run
rein-aharness claim-loop
# Install user service: ./deploy/scripts/install-claim-loop-user.sh
# Docs: docs/ops-run-claim-loop.md

# A queued harness_profile_ref is authoritative: the claim worker builds a
# Glas ExecutionRequest and invokes its resolver/gateway. Install the sibling
# runtime packages in the claim-loop venv:
pip install -e ../sand-boxer -e ../glas-harness

# Legacy: issue-core tickets (external / Forgejo projection only — not FI/Binky ops)
export ISSUE_CORE_URL=http://127.0.0.1:8765 ISSUE_CORE_API_KEY=…
rein-aharness poll --source=issue-core
rein-aharness run --from-issue-core

# Deterministic mailbox scan (no LLM session)
rein-aharness mail-scan --target-repo ~/binky-control

# Mail triage via llm-connect HTTP (OpenRouter behind the service; no Claude CLI)
export LLM_CONNECT_URL=http://llm-connect.activity-core.svc.cluster.local:8080
rein-aharness mail-triage --target-repo ~/binky-control

# Daily and weekly briefs via llm-connect (structured JSON → deterministic markdown)
rein-aharness brief-daily --target-repo ~/binky-control
rein-aharness brief-weekly --target-repo ~/binky-control

# Railiance packaging smoke (commit + optional push + hub; no Claude required)
rein-aharness smoke --work-dir ~/work/executor-sandbox
```

Railiance deploy: [deploy/README.md](deploy/README.md).

Each `run` resolves the agent's tool profile + budget from the target
repo's instance manifest (default `green-commit-only`), loads a persona
bundle (`kaizen-agentic schedule prepare`), runs a bounded agentic
session (cwd-pinned, hard allow-list, never pushes), verifies a local
commit, writes `.kaizen/metrics`, and posts a State Hub progress event.
When invoked through `glas-harness`, the explicit versioned Glas profile may
override model, tool profile, and budget for that bounded run. Workforce and
activity consumers should reference the Glas profile rather than encode these
rein CLI details in an assignment.

Instance manifest contract: [docs/instance-manifest.md](docs/instance-manifest.md).
Example: [examples/schedule.harness.yml](examples/schedule.harness.yml).

Tests: `PYTHONPATH=".:$HOME/llm-connect" python3 -m pytest tests/ -q`