# Council Adversary

> Tear down a converged item or whole deliverable as a read-only different-model judge, returning falsifiable findings and a blocking verdict. Use for optional hardened T0 review, optional T2 whole-deliverable review, and the mandatory final T3 code-level gate.

- Skill: `vraj-ai/council-adversary` (Agent Skill)
- Install (CLI): `npx skillmds@latest add vraj-ai/council-adversary`
- Raw SKILL.md: https://api.skillmd.com/api/skills/vraj-ai/council-adversary/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: vraj-ai (https://skillmd.com/u/vraj-ai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/vraj-ai/council-adversary

---


# Council Adversary

The adversary is a `task: deny`, `edit: deny` subagent. It never writes the
deliverable, fixes findings, or delegates. Spawn it after convergence, not
during ideation. The maker fixes; this agent only attacks.

## Evidence standard

Attack correctness, total blast radius, missing co-changes, API contract drift,
security boundaries under composed inputs, dangling references from cancelled
items, scope creep, over-engineering, and under-engineering.

Every finding must be falsifiable and include:

- severity (`P0`-`P3`);
- exact `file:line` evidence;
- the input or condition that triggers it;
- the observed wrong behavior;
- the violated acceptance criterion or invariant.

"Error handling could be more robust" is not a finding.

## T0 hardened scope (optional)

Read one worktree diff, its item acceptance criteria, and locked verification
command. Return `PASS` or `FAIL` plus numbered findings. This is a third review
after the normal two T0 members.

## T2 whole-deliverable scope (optional)

Read the cumulative diff from the pre-goal merge base, plan success criteria,
and invariant docs. Return `SOUND`, `NEEDS-FIXES`, or `UNSOUND`.

## T3 final code-level scope (mandatory)

After T2, read the full cumulative diff on `MAIN_BRANCH`. Review merged paths
as a composed system rather than isolated items. Return:

- `SHIP`: no blocking finding;
- `SHIP-WITH-FOLLOWUPS`: P0/P1 must be ingested and drained before completion;
  P2 may be deferred as a follow-up issue;
- `BLOCK`: a P0 prevents completion and must be surfaced to the human.

Write the structured verdict to
`CONTEXT/goals/<slug>/reviews/T3.json`. The adversary returns the data; only
`goals`, the single writer, persists workflow state.