What to include when handing a bug from chat to a coding agent

A bug handoff from chat to a coding agent should include the exact symptom, reproduction path, environment, failed attempts, and the smallest safe next step.

Chat is useful for reasoning, but code changes need concrete input. A vague handoff such as “fix the login issue” forces the coding agent to rediscover context. A stronger handoff says what the user did, what happened, what should have happened, which command or page reproduces it, and which files or modules are likely involved. This reduces unnecessary exploration and lowers the risk of editing the wrong layer.

Include the latest error output, but trim noise. A full log dump can hide the relevant line. Pair the log with a short statement: “fails after token refresh,” “only happens on mobile width,” or “test passes locally but fails in CI.” If the issue is intermittent, include timing, frequency, and what changed recently.

List failed attempts separately. This prevents the next tool from repeating cache clears, dependency installs, or config changes that already failed. If an attempt partially helped, say how. Partial evidence is useful because it narrows the search.

Finally, name the next safe action. That might be “inspect auth middleware,” “write a failing test,” “compare request payloads,” or “check environment variable loading.” The handoff should not dictate a fix before evidence supports it, but it should give the coding agent a defensible starting point.

The practical goal is continuity. The second tool should inherit the problem, not the confusion.

What to include when handing a bug from chat to a coding agent

// COMMENTS

ON THIS PAGE