Skip to main content

Architecture

Better Harness architecture: host integration, three independent evidence agents, unified analysis by one lead agent, findings, host outputs, and repair

The architecture keeps the three evidence domains independent until unified analysis by the lead agent. Every result retains a visible evidence source, owner, and validation route.

Feedforward and feedback

Better Harness uses a feedforward-and-feedback loop that combines guidance available before work starts with signals available after the agent acts:

  • Feedforward guidesAGENTS.md, specs, Skills, and acceptance criteria steer the agent before it acts.
  • Feedback sensors — linters, tests, Hooks, and review agents observe results and help the agent self-correct.

Where things live

DirectoryRole
skills/Repeatable agent workflows (start at better-harness)
models/Evaluation models; default first, advanced second
references/Prose guidance, loaded on demand
templates/Report skeletons, output modes, and styles
hooks/Change-time enforcement
scripts/Capability-owned CLIs

What runs under the hood

CapabilityCLIJob
Quickstartbetter-harness reportGather evidence and hand off to the skill
Readiness analysis/better-harness skillSynthesize the evidence-backed report
Project evidencebetter-harness core-change-watchProject, history, core-path, and diff signals
Change confidencehooks/git-scripts/blast-radiusSymbol-graph blast radius of a change
Dependency governancebetter-harness dependency-governanceUpdate automation, audit, stale-dep signals
Session evidencebetter-harness session-analysisNormalize Qoder, Codex, Claude, or Cursor session behavior
Agent assetsbetter-harness coding-agent-practices inventoryInventory configured agent surfaces
Guardrailshooks/, scripts/agent-guardrailsSecret scanning and lifecycle checks

Source of truth

Directory routing, template ownership, and canonical-owner rules live in docs/ARCHITECTURE.md.