agents-harness

Declarative scheduled flow runner, per-request agent loop, and job kernel for local agent services.
Zero heavy dependencies (Python standard library only). No Celery, no Redis, no cron daemons.

Architecture Principles

• Per-request loop: The agent is a reconstructed trace (python -m runner.loop). SKILL/MCP/A2A entries live in unner/modules/ as Cordis verbs (when=on_request or later). Scheduled care jobs stay in unner/schedules/.
• MCP server support: Optional extra pip install -e ".[mcp]" with FastMCP support (python -m runner.mcp_server). The executor stays stdlib-only.
• Strict Layering (Cordis Principle): The executor owns no state, no domain store, and no semantics. It strictly invokes verified CLIs and inspects exit codes.
• Job kernel: Declarative catalog JSON is the alphabet. Checked, mixed, and reduced deterministically (python -m runner.kernel).
• Identity directory: Decoupled alias, user, and session identifiers (~/.agents/identity.json).

Commands

`ash

Validate manifests

python -m runner.executor --validate

Run scheduled flows

python -m runner.executor --all python -m runner.executor --run

Loop & execution

python -m runner.loop --list-modules python -m runner.loop --list-tools python -m runner.loop --list-providers python -m runner.loop --assemble-only --user --message "hello"

FastMCP server

python -m runner.mcp_server

Kernel

python -m runner.kernel --alphabet `

License

MIT License. Copyright (c) 2026 Felix Kempf.