Harden supervisor evidence and loop stops#57
Conversation
There was a problem hiding this comment.
Code Review
This pull request updates the supervisor loop and dynamic loop to handle child help passthroughs, restrict loop-owned options from being passed after --, and preserve specific exit codes (such as exit code 3 for inspection-required stops). It also refines transaction hash extraction and validation during classification to catch mismatched hashes as malformed evidence. Feedback suggests refactoring spawnSupervisorHelp to use 'pipe' for stdio and capture output safely, preventing potential Node.js runtime crashes when custom streams are passed to spawnSync in test environments.
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request introduces stricter validation of transaction hashes in actor evidence, ensuring that mismatched transaction hashes within a single record are rejected as malformed evidence. It also updates the supervisor loop and dynamic loop scripts to properly handle child help passthroughs ('--help' and '-h'), prevent misplaced loop-owned options after the '--' separator, and preserve the inspection-required exit code (3) when stopping due to 'max_runs' or 'stable_no_progress'. Extensive unit tests have been added to verify these behaviors. There are no review comments, so I have no feedback to provide.
|
LGTM Phroi %131 |
Why
Supervisor artifacts drive live validation and stop decisions. Tx-hash evidence can appear in top-level records and nested error/skip fields, and loop stop reasons need clear operator semantics.
Changes