Skip to content

Retro agent should not file issues for findings humans explicitly declined during review #2957

Description

@fullsend-ai-retro

What happened

On PR #12 in fullsend-ai/agents, ggallen explicitly commented (14:29 UTC) that the customer-research finding was declined: "optional/graceful." Despite this clear human signal, the retro agent filed issue #13 as a bug for the same finding. This triggered a triage → code → fix PR cascade that was entirely unnecessary, wasting agent compute and requiring human intervention to clean up.

What could go better

The retro agent analyzes PR review findings and files follow-up issues for unresolved ones. However, it does not distinguish between findings that were unresolved due to oversight and findings that were explicitly declined by a human reviewer. When a human comments on a review finding explaining why they chose not to address it, that is a deliberate decision, not an oversight. The retro agent should treat human-declined findings differently from silently ignored ones. This is related to but distinct from existing issue #2502 (proportionality assessment) and #2949 (higher bar for low-severity findings) — those address severity thresholds, while this addresses respecting explicit human decisions. Confidence: high — the human signal was unambiguous.

Proposed change

Update the retro agent's analysis logic (in the retro-analysis skill or the retro agent definition) to check whether unresolved review findings were explicitly addressed by a human commenter. When a human comment on the PR acknowledges a finding and provides a reason for not fixing it (e.g., "intentional", "by design", "declined", "not a bug", "optional"), the retro agent should either skip filing an issue for that finding or, if filing, note the human's rationale and label it as requires-manual-review rather than bug + ready-to-code. The key heuristic: if a human explicitly responded to a review finding with a reason for not addressing it, that finding should not become an automated follow-up issue.

Validation criteria

On the next 5 PRs where a human explicitly declines a review finding with a stated reason, the retro agent should not file a follow-up issue for that finding (or should label it appropriately if filed). Zero false-positive issues from human-declined findings over a 30-day window.


Generated by retro agent from fullsend-ai/agents#13

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    Status
    Todo

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions