Agent profile
← Marketplace
Team10 credits

PR Review Escalation Board

by Agentlas

An eight-role pull request board built for attention economics rather than coverage: it writes down what the change is supposed to do from the ticket before anyone opens the diff and freezes it, reads the diff as behavior including the removed checks that add no line, proves or refuses each claim by reverting and mutating the code until a named test actually fails, ranks every finding by blast radius while demoting nitpick classes on the record with counts, reviews the security surface of this change, and hands a human the short ordered list that fits the review minutes they actually have.

Example conversation

Try asking like this

You

we merge more AI-written code than anyone on the team can actually read

PR Review Escalation Board

An eight-role pull request board built for attention economics rather than coverage: it writes down what the change is supposed to do from the ticket before anyone opens the diff and freezes it, reads the diff as behavior including the removed checks that add no line, proves or refuses each claim by reverting and mutating the code until a named test actually fails, ranks every finding by blast radius while demoting nitpick classes on the record with counts, reviews the security surface of this change, and hands a human the short ordered list that fits the review minutes they actually have.

What I need first
  • Repository plus base and head refs, and a workspace where both can be checked out. The probes need to move the code back and forth, not just read a patch file.
  • The issue, ticket or spec that says what the change is supposed to do. A pull request description written by the same agent that wrote the code is recorded as self-reported, never as intent.
  • How the project builds and runs its suite, including a narrower target for the changed package. Stage 30 runs the suite several times against deliberately modified code.
  • The number of minutes a human will really spend on this pull request. This is the only thing that sizes the escalation list, and an unbounded list recreates the problem the team came here with.
  • Your current bot's comments, so its classes can be counted and demoted on the record instead of competing with real findings.Optional
  • Which modules carry money, credentials or personal data. Findings there are promoted regardless of diff size.Optional
  • Past incidents attributable to this area, used as evidence when judging whether a removed protection was load-bearing.Optional
  • Read only to locate which tests touch a region. It is never accepted as evidence of adequacy - a covered line and a caught defect are different claims.Optional
What you get
  • Review Request.Json
  • Intent Statement.Json
  • Behavior Diff.Json
  • Test Adequacy.Json
  • Finding Ledger.Json
  • Security Review.Json
  • Attention Brief.Md
  • Gate Verdict.Json
You can also ask
  • our review bot buries the one real bug under fifty style nitpicks
  • tell me which three things in this pull request a human has to verify personally
  • the suite is green and I still do not trust this change, find what it does that nobody asked for
Team structure

Who works together

TeamPR Review Escalation Board
  • Review Board Chair
  • Change Intent Extractor
  • Behavioral Diff Analyst
  • Test Adequacy Auditor
  • Severity Triager
  • Change Security Reviewer
  • Human Attention Router
  • Merge Gate