Enable a harness

A harness is an AI coding tool on your machine — a signal source. Trove auto-detects the harnesses you have installed (17 are supported, from Claude Code and Antigravity CLI to Cursor, Cline, Aider, and GitHub Copilot CLI) and can turn telemetry on for each one with a single toggle.

Enabling a harness does one of two things depending on how that tool reports data:

• Native OTel harnesses (Claude Code, Antigravity CLI, Codex, Qwen, OpenCode, Cursor IDE) already speak OTLP. Trove just flips the right env/config flags so they point at the local collector.
• Best-effort harnesses (Cline, Aider, GitHub Copilot CLI) don’t emit OTel natively. Trove installs a lightweight watcher (a filesystem monitor) and/or a shell-rc wrapper that derives equivalent OTLP records from the tool’s on-disk logs. Those derived records answer the same queries as their native peers.

Either way the change is atomic, idempotent, and reversible.

Enable it

1. Open Trove and go to the Harnesses tab. Each detected tool appears as a row with a coverage badge and an enable/disable toggle.

   

2. Find the harness you want and flip its toggle on.

3. Trove writes the patch to that harness’s config file (or shell rc, for wrapper-based harnesses). The write is atomic — temp file, fsync, rename — and the original is backed up first. Re-enabling an already-enabled harness is a no-op.

Tip

Connect a backend first. With a destination configured, the moment a harness starts emitting you’ll see its data land instead of piling up locally.

What the managed-region patch looks like

Trove never scatters edits through your config. Everything it writes goes inside a single sentinel-bracketed managed region — a clearly marked block fenced by Trove sentinel comments. Revert finds that exact block and removes it byte-for-byte, so your file returns to its pre-Trove state with nothing left behind.

Conceptually, an edit to a harness config looks like this:

{
  "your": "existing settings stay untouched",

  // >>> TROVE MANAGED REGION — do not edit by hand <<<
  "telemetry": {
    "enabled": true,
    "otlpEndpoint": "http://127.0.0.1:4317",
    "otlpProtocol": "grpc",
  },
  // <<< END TROVE MANAGED REGION >>>
}

For a wrapper-based harness the managed region lands in your shell rc (for example ~/.zshrc) instead, defining a shell function that re-execs the real binary and logs one structured line per invocation, which the watcher turns into OTLP.

Caution

Don’t hand-edit inside the managed block. If Trove detects that the region was modified by hand, it refuses to silently overwrite your change — it surfaces the conflict instead of clobbering your edit. Settings outside the block are always yours and are never touched.

Confirm OTLP is flowing

1. Run a single turn through the harness — one prompt, one response.

   Claude Code
   bash claude -p "say hi in one word"
   Antigravity CLI
   bash agy -p "say hi in one word"
   Any harness
   Just use the tool normally for one quick interaction.

2. Open the Logs tab and watch the live collector tail. You should see an OTLP batch arrive tagged with this harness’s harness.id resource attribute.

3. Switch to the Overview tab. The data-flow chart shows the harness as an active source feeding the collector, and the Platforms tab’s health pill for your backend should pulse green.

4. Confirm it landed at the destination by running a query in your backend filtered on harness.id (for example harness.id = "claude-code"). One recent row means the full path works end to end.

Note

Native-OTel harnesses read their telemetry config at process start, so a tool that was already running may need to be restarted to pick up the new flags. Wrapper-based harnesses pick up the new shell function in any new shell session.

Disable and revert

Turning a harness back off is the inverse of enabling it, and it’s clean:

1. On the Harnesses tab, flip the harness’s toggle off.

2. Trove locates the sentinel-bracketed managed region and removes exactly what it wrote — the file is restored byte-for-byte to its pre-enable state. No orphaned env vars, no half-applied config, no CLI that keeps POSTing to localhost after you thought you were done.

3. For watcher/wrapper harnesses, Trove also stops the filesystem watcher and removes the shell-function block from your rc file.

This reversibility is the whole point: every “Enable” is something you can take back with one click, which is exactly what makes it safe to try Trove across a whole team.

Connect your first backend Pick a destination and verify the health pill turns green.

Customize mappings Shape this harness's raw signals into the Tier A schema.

Harnesses reference The full list of supported tools and their coverage tiers.

Add identity tags Tag signals with user.name / user.email for team dashboards.