Bug Predict errors

Common error signatures

Errors in bug predict fall into three categories: failures during workflow execution, failures during report formatting, and failures in the subagent coordination layer.

Workflow execution (BugPredictionWorkflow.execute)

• ValueError — the path argument in the task prompt template is missing or resolves to a location the orchestrator cannot access.
• RuntimeError — one or more of the three subagents (pattern-scanner, risk-correlator, prevention-advisor) did not return a result before the workflow completed.

Report formatting (format_bug_predict_report)

• KeyError — result or input_data is missing a field the formatter expects (for example, severity, findings, or risk_score).
• TypeError — result or input_data was passed as None or an unexpected type instead of dict.

CLI entry point (main)

• SystemExit with a non-zero code — the path argument was not supplied or could not be resolved before BugPredictionWorkflow was instantiated.

Where errors originate

• BugPredictionWorkflow.__init__ in workflows/bug_predict.py — validates constructor arguments, including system_prompt_suffix, before the workflow starts. A bad value here prevents any scanning from running.
• BugPredictionWorkflow.execute in workflows/bug_predict.py — coordinates pattern-scanner, risk-correlator, and prevention-advisor. If subagent synthesis fails, the error surfaces here as a WorkflowResult with no findings or as an unhandled exception.
• format_bug_predict_report in workflows/bug_predict_report.py — converts the raw result dict and input_data dict into the human-readable report. Malformed or incomplete dicts from a partial execute run will cause this function to fail.
• main in workflows/bug_predict_report.py — the CLI entry point. Errors here typically mean the workflow never ran, not that it ran and produced bad output.

How to diagnose

1. Identify which layer failed. Check whether the traceback points to bug_predict.py (workflow or subagent coordination) or bug_predict_report.py (formatting or CLI). The fix differs by layer.

2. Check the result dict shape. If the error is a KeyError or TypeError inside format_bug_predict_report, inspect the WorkflowResult returned by execute. A partial run — for example, one where risk-correlator or prevention-advisor did not finish — produces an incomplete dict that the formatter cannot process.

3. Check the path argument. _TASK_PROMPT_TEMPLATE requires a {path} value. If execute was called without a resolvable path, the orchestrator prompt is malformed and all three subagents receive bad input. Confirm the path exists before calling execute.

4. Inspect false-positive filter state. If the scan runs but returns no findings on code you expect to be flagged, check whether the matched lines contain any _INTENTIONAL_KEYWORDS (fallback, ignore, optional, best effort, graceful, intentional) or a # INTENTIONAL: / # noqa: BLE001 marker, which the scanner suppresses automatically.

5. Verify test files are not contaminating results. Files matching _SCANNER_TEST_PATTERNS (test_bug_predict, test_scanner, test_security_scan) are excluded from scanning. If your production code path shares a name with one of these patterns, it will be silently skipped.

Source files

• src/attune/workflows/bug_predict.py
• src/attune/workflows/bug_predict_report.py

Tags: bugs, prediction, scanning