HarnessFacts
Open conformance suite · v0.1

Tested facts about
AI coding agents.

What does your coding agent actually support? Small, reproducible tests for Codex, Antigravity CLI, OpenCode and more.

No rankings. No synthetic scores.
Just documented and observed behavior.

● ● ●   reproducible by design
$ harnessfacts run codex instructions.nested

01   Create a temporary fixture
02   Run the agent
03   Check the exact output
04   Save the evidence

A fact you can inspect. A test you can repeat.
$ harnessfacts compare codex opencode
6Agent adapters
12Executable tests
70Completed observations
100%Open methodology

The compatibility notebook

Every observation has a version, an environment, and evidence.

Observed results and official documentation are separate evidence. Missing data remains unknown.
CapabilityOpenAI CodexAntigravity CLIOpenCodeClaude Code
Root instructions✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: unknown
Nested instructions✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: unknown
Instruction precedence✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: unknown
MCP stdio✓ OBSERVEDv0.153.4 · linux x64Documented: yes ↗✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: unknown
MCP HTTP✓ OBSERVEDv0.153.4 · linux x64Documented: yes ↗✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: unknown
Headless execution✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: yes ↗? UNKNOWNDocumented: yes ↗
Exit semantics✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: unknown
Structured output✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: yes ↗
Skill discovery✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: yes ↗? UNKNOWNDocumented: unknown
Hook execution? UNKNOWNDocumented: unknown? UNKNOWNDocumented: unknown? UNKNOWNDocumented: unknown? UNKNOWNDocumented: unknown
Git worktrees✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: unknown
Repository boundary✓ OBSERVEDv0.153.4 · linux x64Documented: unknown✓ OBSERVEDv1.1.27 · linux x64Documented: unknown✓ OBSERVEDv1.18.29 · linux x64Documented: unknown? UNKNOWNDocumented: unknown
✓ Observed in a completed run× Expected behavior not observed? Insufficient evidenceHow to read these results →
01 / REPRODUCE

Small tests. Inspectable results.

See the fixture, prompt, and exact condition behind every capability.

Browse the suite →
02 / UNDERSTAND

Keep uncertainty visible.

A documented feature is a claim. An observed result records one execution.

Read the methodology →
03 / CONTRIBUTE

One result makes a difference.

Run a test on your machine, add an adapter, or help verify another platform.

Contribute a result →