Agent Codex
Devil's advocate against a freshly formulated implementation plan, before the user approves it. Attacks the approach with codebase evidence — concrete failure scenarios and cheaper alternatives — never vague doubt. Invoke in /ticket-pick step 3, after the plan is drafted and before the Plan gate, so the user judges…
Agent Codex
Devil's advocate against a ticket's implementation as built — run every round of the pick loop, not once. Attacks the code that now exists with evidence — hidden coupling it introduced, a cheaper route it should have taken, an irreversible or load-bearing step, or a route the plan got wrong. Read-only; returns a…
Agent Codex
Read-only review of a ticket's implementation diff before it transitions to review. Checks the diff against the approved plan, architecture invariants, and project conventions. Invoke in /ticket-pick's implementation loop (step 5.5 agent checks) — after verification passes, every round — before the review transition…
Agent Codex
Proposes simplifications to a ticket's implementation diff — removing accidental complexity, dead abstraction, and speculative generality — without changing behavior. Read-only; returns proposals the session weighs in its round evaluation and folds into the loop's work-list (no user gate). Invoke every round of…
Agent Codex
Derives a ticket's non-functional requirements while it is still being written — as measurable statements that each carry their own verification. Invoke in /ticket-new step 2 (alongside the research agents, before the step 2.5 grilling) and on /ticket-refine's resume path, so unresolved requirements become grilling…
Agent Codex
Judges whether the tests accompanying a ticket's diff would actually fail if the behavioral change were reverted or broken. Catches assertion-free tests, tests that only exercise mocks, and untested branches. Invoke in /ticket-pick's implementation loop (step 5.5 agent checks) after tests pass, every round — a green…
Agent
Six read-only subagents, all tools: Read, Grep, Glob, Bash — none of them edit anything, and none of them gate you directly. Each returns a verdict that the calling session weighs; only the main session ever acts on a finding. Alongside the research agents you register yourself (see Getting started § Step 0), these…
Agent Claude Code
stub for the scaffolded example.
Agent Claude Code
stub for the scaffolded example.
Agent
stub for fixture.