Verification

Nyzhi can automatically detect and run your project’s build, test, and lint checks. The verification system provides structured evidence of pass/fail status, making it easy for the agent to confirm its changes are correct. Use it from the TUI, as a tool, via hooks, or in autopilot QA phases.

Auto-detection

When you run /verify or the agent calls the verify tool, Nyzhi scans the project root to determine the project type and applicable checks:

Rust

Node.js

Go

Python

Detection is based on the presence of marker files (Cargo.toml, package.json, go.mod, pyproject.toml, etc.).

Evidence

Each check produces structured evidence:

• Build / Test / Lint — The exact command run, exit code, stdout, stderr, timestamp, and elapsed time
• passed() — true if exit_code == 0
• is_fresh(max_age) — true if the evidence is recent enough

Verify report

Multiple checks produce a VerifyReport:

Verification Report:
  ✓ Build (cargo check) — 2.3s
  ✓ Test (cargo test) — 8.1s
  ✗ Lint (cargo clippy --all -- -D warnings) — 1.2s
    error: unused variable `x`

2/3 checks passed

• all_passed() — true only if every check succeeded
• summary() — human-readable report with timing and failure details

Usage

In the TUI

/verify          # show detected checks and run them

As a tool

The agent can call verify as a tool to check its work:

Tool: verify
Result:
  Build: ✓ (2.3s)
  Test: ✓ (8.1s)
  Lint: ✓ (1.2s)
  All checks passed.

In hooks

Run verification after each turn:

[[agent.hooks]]
event = "after_turn"
command = "cargo test"
timeout = 120

In persist mode

/persist activates a verify/fix loop: the agent runs checks, identifies failures, fixes them, and re-runs checks until everything passes.

Custom checks

While auto-detection covers common project types, you can run arbitrary commands through hooks:

[[agent.hooks]]
event = "after_turn"
command = "make lint && make test"
timeout = 180

Check kinds