Harmony

Cross-platform drift detection for real multi-agent repos.

Harmony is the Nerviq layer that asks a hard question most tooling ignores: when the same repository is used by multiple agent surfaces, are those surfaces actually aligned or slowly diverging into contradictory operating systems?

Drift detectionSync guidanceGovernance summary

What Harmony Does

Harmony turns a repo with multiple agent configurations into one comparable model. It treats instructions, trust posture, MCP coverage, and governance notes as a shared system instead of isolated files.

Canonical model

A normalized view of instructions, MCP servers, trust posture, and governance across every detected platform.

Command surfaces

Audit, sync, drift, advise, watch, and governance all target the same model.

Overwrite-by-default policy

Harmony is designed to preserve platform-native files and add the minimum needed to close drift.

Command Family

Harmony is not one command. It is a set of operating views over the same cross-platform state.

Command	Purpose
`harmony-audit`	Scan all active platforms in a repo and score the drift between them.
`harmony-sync`	Generate or patch missing cross-platform assets based on the canonical model.
`harmony-drift`	Explain where instructions, trust settings, MCP servers, or governance posture diverge.
`harmony-advise`	Prioritize the smallest set of changes that reduces the most cross-platform inconsistency.
`harmony-watch`	Continuously monitor a repo and surface new drift when files or configs change.
`harmony-governance`	Render a governance-facing summary of trust, MCP, and surface alignment.

How It Works

Harmony follows a predictable loop so teams can reason about changes before they synchronize anything.

Normalize

1. Build the canonical model

Nerviq reads every active platform surface it can find, then normalizes instructions, MCP server lists, trust posture, and governance notes into one comparable structure.

Compare

2. Detect drift

Harmony looks for gaps such as missing MCP servers, contradictory trust settings, or platforms that mention workflows no other platform can see.

Patch

3. Sync with preservation

When you choose to act, Harmony adds or patches the smallest set of files needed to reduce drift. Existing platform-owned content is meant to stay intact.

Measure

4. Score the result

After the sync plan, Harmony reports an alignment score so teams can see whether they are converging on a coherent multi-agent setup or still operating with fragmented rules.

Example Output

A Harmony report is designed to read like an operator briefing rather than raw diff noise.

bash

nerviq harmony-audit --format markdown

# Example output
Alignment score: 82
Active platforms: claude, codex, cursor

Drift:
- MCP server "context7" exists on Claude and Codex but is missing from Cursor
- Cursor has privacy guidance not mirrored in CLAUDE.md or AGENTS.md
- Codex review workflow is documented, but Claude lacks a matching verification note

Recommended next action:
- Add shared MCP coverage and one repo-wide verification block, then rerun Harmony

Why Teams Use It

Harmony is what keeps a repository from splitting into eight incompatible AI operating models.

Operator benefit

Less hidden drift

Teams can see when one platform silently stops matching the rest, instead of discovering the mismatch through inconsistent agent behavior weeks later.

Governance benefit

Higher-confidence rollouts

Security, review, and MCP posture can be discussed as one model. That makes platform expansion safer because every new surface inherits the same alignment checks.