Duplicate Error Hashing

GitAuto hashes CI error logs after each test run and compares the hash against previous attempts. If the same error hash appears twice, the agent knows its fix did not work and should try a different approach or give up.

Why This Exists

Without error hashing, the agent can make the same incorrect fix repeatedly, creating 10+ commits that all fail in the exact same way. Each commit triggers a full CI run, and the agent reads the same error log, generates the same wrong fix, commits it, and loops. Error hashing breaks this cycle by detecting when the agent is stuck.

How It Works

After each CI run, GitAuto extracts the relevant error portion of the logs, normalizes it (removing timestamps, line numbers, and other volatile content), and computes a hash. This hash is stored alongside the iteration number. Before starting a new fix attempt, the agent checks whether the current error hash matches any previous hash. If it finds a duplicate, it knows the previous fix was ineffective and can either try a fundamentally different approach or stop and report the issue.

Related Features

• Infrastructure Failure Detection - distinguishes infrastructure errors from code errors before hashing