code-challenger

The loop-time sibling of challenger — attacks the route the code as built actually took, not the paper plan. Runs every round, on whatever the diff has become by then.

Generated for your project

This page documents the agent's contract, which never changes. /ticket:init regenerates its role, focus and knowledge regions for your stack — the coupling, failure scenarios and cheaper routes typical of it, each with its source. It is always on; project advisors (review.advisors) can join it every round. See Generated agents.

Triggered by	/ticket:pick step 5.5, every loop round, alongside code-reviewer — advisory, always on, not configurable
Also usable	Standalone, on any diff
Tools	Read, Grep, Glob, Bash

Flow

flowchart LR
    In1["Ticket + plan"] --> A["code-challenger"]
    In2["Diff base + current HEAD"] --> A
    In3["Round ≥ 2: prior findings +<br/>change summary"] --> A
    A --> Out["Verdict + up to 3 challenges"]
    Out --> Eval["Step 5.7 evaluate —<br/>sound findings fold into<br/>next round's work-list"]
    Out -.->|"ROUTE-WRONG"| Replan(["Can trigger a re-plan"])

Verdicts

Verdict	Meaning
CODE STANDS	No material weakness in the route taken
WEAKNESSES	Named concerns — foldable into the next round's work-list
CHEAPER ROUTE	A less costly implementation exists
ROUTE-WRONG	The route taken is fundamentally wrong — can send the loop back to re-plan
ASSUMPTION-BROKEN	A load-bearing assumption no longer holds

Gates directly? No — advisory. The step 5.7 evaluation weighs its findings and can fold sound ones into the next round's work-list; only a ROUTE-WRONG verdict carries enough weight to trigger a re-plan on its own.

See also

• challenger — the plan-stage counterpart, run once before implementation starts