Skip to content

Review agent should cross-reference triage findings when reviewing PRs linked to triaged issues #1351

@fullsend-ai-retro

Description

@fullsend-ai-retro

What happened

The triage agent on issue #59 flagged a specific implementation concern: because init_schema() uses INSERT OR IGNORE, existing users who already have a database won't pick up the new default for hide_stale_sessions. The triage bot recommended verifying whether a migration step or fallback logic is needed. When PR #60 was opened, it did not address this concern — it only changed DEFAULT_SETTINGS without adding migration logic. The review agent approved the PR without flagging the gap. The triage bot also proposed two specific test cases (verifying the default is '1' and verifying user overrides are preserved), neither of which appeared in the PR. The review did not note their absence.

What could go better

The review agent should incorporate context from the originating issue's triage summary when one exists. The triage agent's implementation notes and proposed test cases represent domain-specific guidance that was generated earlier in the pipeline — ignoring them at review time means the pipeline's own earlier analysis gets wasted. I'm fairly confident this is a real gap: the triage agent explicitly called out a migration risk, and the review agent had no mechanism to surface or validate that concern. The doc findings the review did catch were lower-impact than the migration issue it missed.

Proposed change

In the review agent's context assembly (likely in agents/review.md or scripts/pre-review.sh), when the PR is linked to an issue that has a triage bot comment (identifiable by the <!-- fullsend:triage-agent --> marker or similar), fetch and include that triage summary in the review agent's context. The review prompt should instruct the agent to cross-check: (1) were implementation concerns from triage addressed? (2) were suggested test cases included or explicitly deemed unnecessary? This could be implemented as a pre-review script step that fetches the linked issue's triage comment and injects it into the agent's context, or as an addition to the review agent's prompt instructing it to check the linked issue.

Validation criteria

On the next 5 PRs linked to triaged issues where the triage bot flagged implementation concerns or suggested test cases, the review agent should reference those concerns in its review output — either confirming they were addressed or flagging them as gaps. Measure by checking whether review comments mention triage-identified risks.


Generated by retro agent from agentshed/seshi#60

Metadata

Metadata

Assignees

No one assigned

    Type

    No type
    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