Agent-harness audit lens (loop/context/policy/tools)

Generated from mcp/patterns.registry.json — do not edit by hand; change the registry and run bun run patterns:build.

Index card — the docs below are the canonical explanation; deep-tier pattern pages additionally carry invariants and emulation notes.

Category: AI · Tier: light · Risk: medium — governs the agent loop that drives external LLM providers

A conceptual audit lens (not a module) names which slice — loop, context, policy, or tools — owns each agent-harness primitive such as compaction, step caps, and approval gating.

When to use: Use when auditing whether an LLM-agent product has the harness primitives it needs and where each one is implemented.

Docs

• docs/blueprint/ai/harness-lens.md (GitHub · GitLab)

Code

• src/lib/server/ai/loop/ (GitHub · GitLab)
• src/lib/server/ai/policy/ (GitHub · GitLab)

Machine-readable record: ai-harness-lens in mcp/patterns.registry.json.