Result Contract

Every step must produce a result marker. This is the protocol between the agent and the bugatti harness.

Result markers

The agent emits one of these at the end of its work for each step:

RESULT OK
RESULT WARN: <message>
RESULT ERROR: <message>

Marker	Meaning	Exit code impact
RESULT OK	Step passed	0
RESULT WARN: <msg>	Passed with caveats	0 (or 1 with --strict-warnings)
RESULT ERROR: <msg>	Step failed	1

How bugatti parses results

Bugatti searches backwards through the agent’s output for the last RESULT line. This means:

• The agent can output free-form reasoning before the marker
• Only the final RESULT line counts
• If no marker is found, it’s a protocol error and the step fails

Timeouts

If the agent doesn’t produce a result within the step timeout (default 300s, configurable per-step or globally), bugatti kills the agent process and records a timeout failure (exit code 4).

Agent logging

The agent can emit structured log lines during execution:

BUGATTI_LOG Created test user with id=42
BUGATTI_LOG Screenshot saved to /tmp/login-page.png

These are:

• Displayed in the console as LOG ........ <message>
• Captured separately from the full transcript in run artifacts
• Useful for tracing what the agent did without reading the full output

Run artifacts

Each run saves artifacts to .bugatti/runs/<run_id>/:

.bugatti/runs/20260403-141523-001-a1b2/
├── metadata.json      # Run config, test file, timestamps
├── report.md          # Human-readable summary
├── transcripts/       # Full agent output per step
├── screenshots/       # Evidence captured by agent
├── logs/              # BUGATTI_LOG entries
└── diagnostics/       # Structured tracing data

The run ID format is YYYYMMDD-HHMMSS-fff-XXXX (timestamp + 4-char suffix for uniqueness).