skills.rest
Priivacy-ai
Priivacy-aiOFFICIAL
Explore all skills by this author →
GitHub

adversarial-squad

Deploys parallel profile-loaded review agents to critique specs, plans, and tasks.

1.6k|156|Updated Oct 9, 2025
Search Tags:#parallel agents#adversarial review#design critique#spec-kitty#Code review#spec-Driven Development#Multi-Agent
One-click install
npx skills add https://github.com/Priivacy-ai/spec-kitty --skill adversarial-squad
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: adversarial-squad
Source: https://github.com/Priivacy-ai/spec-kitty/tree/main/src/charter/offering/skills/adversarial-squad
Command: npx skills add https://github.com/Priivacy-ai/spec-kitty --skill adversarial-squad

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Single-reviewer blind spots let scope errors, fakeable definitions of done, and architectural flaws slip through spec-driven development. This Skill dispatches a bounded squad of independent review lenses so convergent findings survive scrutiny one reviewer would miss.

Core Features & Use Cases

  • Point-cut deployment: Run a squad after specify, plan, or tasks, before merge, or on an ad-hoc decision such as delete-vs-migrate.
  • Profile-loaded delegates: Each of 3-4 agents loads a distinct doctrine profile (architecture, debugging, anti-laziness review, decomposition) via the spec-kitty CLI before reviewing.
  • Structured verdicts: Findings returned as severity-tagged file:line issues with evidence, concessions, and second-opinion adjudication on divergence.
  • Use Case: After generating tasks for eight work packages, run a post-tasks anti-laziness pass to catch fakeable acceptance criteria before implementation begins.

Quick Start

Ask the agent to run adversarial-squad as a post-tasks anti-laziness review on your current work packages.

Frequently Asked Questions about adversarial-squad

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an adversarial review on my spec-kitty tasks?▼

Invoke the skill by name with a point-cut and question, such as a post-tasks anti-laziness pass on specific work packages. The orchestrator dispatches 3-4 profile-loaded delegates in parallel and synthesizes their structured findings.

What is a multi-lens code or design review squad?▼

It is a bounded group of 3-4 independent review agents, each loaded with a distinct doctrine profile covering architecture, debugging evidence, anti-laziness review, or decomposition. Independent lenses converge on findings a single reviewer would miss.

When should I deploy an adversarial squad in spec-driven development?▼

Deploy at high-leverage point-cuts: after specify for pre-spec investigation, after plan for brownfield checks, after tasks for anti-laziness review, before merge for architectural gates, or for ad-hoc decisions like delete-vs-migrate.

Can the adversarial squad block or gate a mission?▼

No. The squad is an optional, charter-activated enrichment technique and is explicitly never a mission gate. It does not handle implementation, spec generation, or direct code editing by the orchestrator.

What happens when squad delegates disagree on a finding?▼

Divergent conclusions on consequential points are adjudicated from the source rather than averaged, or resolved by dispatching one focused second-opinion delegate. Irreconcilable disagreements escalate to the human with both positions.

Related Skills

Explore More Software Engineering Skills→
SpillwaveSolutionsSpillwaveSolutions
91

sdd

Guide Spec-Driven Development workflows with GitHub's Spec-Kit.

Official
Advanced
kimprobablykimprobably

spec-kitty-mission-review

Trace merged Spec Kitty mission implementations from spec artifacts to tests and code diffs.

Community
Advanced
jitraponjitrapon
11

address-review

Evaluate and resolve open findings in .claude/REVIEW_ADVERSARIAL.md.

Community
Advanced
alirezarezvanialirezarezvani
25.3k

hivemind

Orchestrate parallel opencode workers from Claude Code with worktree isolation and consensus aggregation.

Community
Advanced