uno-reverse
Radical simplification, red-teaming, and Occam's Razor devil's advocate for software architecture, PRDs, agent workflows, and technical designs. Challenges feature creep, speculative abstractions, and bloated specifications by proposing minimum viable primitives that deliver 90% of value with 10% of moving parts. Prefers impartial dedicated subagent delegation to prevent context contamination and confirmation bias. Activate when reviewing complex technical proposals, pruning bloated architectures, red-teaming design docs, eliminating speculative features, or seeking the simplest possible path to production.
Daniela Petruzalek (daniela@danicat.dev) · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add danicat/skills --skill uno-reverse技能说明
name: uno-reverse description: > Radical simplification, red-teaming, and Occam's Razor devil's advocate for software architecture, PRDs, agent workflows, and technical designs. Challenges feature creep, speculative abstractions, and bloated specifications by proposing minimum viable primitives that deliver 90% of value with 10% of moving parts. Prefers impartial dedicated subagent delegation to prevent context contamination and confirmation bias. Activate when reviewing complex technical proposals, pruning bloated architectures, red-teaming design docs, eliminating speculative features, or seeking the simplest possible path to production. license: Apache-2.0 compatibility: agents: - Antigravity metadata: category: agents tags: "simplification, red-team, architecture, occams-razor, minimalism, review, refactoring" author: Daniela Petruzalek (daniela@danicat.dev) version: "0.1.2" catalog: https://skills.danicat.dev
Uno-Reverse: Radical Simplification & Red-Teaming
"Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away." — Antoine de Saint-Exupéry
The uno-reverse skill provides a rigorous contrarian simplification and red-teaming framework. While standard engineering processes naturally drift toward feature accretion, defensive layering, and speculative generalization, uno-reverse forces the opposite trajectory: aggressive subtraction, primitive collapsing, and minimum viable execution.
🎭 Impartial Review: Dedicated Subagent Delegation (Default)
To guarantee the integrity and rigor of an Occam's Razor red-team assessment, uno-reverse reviews must be conducted by a dedicated, isolated subagent rather than within the invoking agent's active conversation context.
1. Rationale: Why In-Context Audits Fail
When an agent that designed, brainstormed, or iteratively implemented a solution attempts to evaluate its own work inline, the review suffers from three critical failure modes:
- Confirmation Bias & Sunk-Cost Defense: An agent that spent multiple conversation turns proposing abstractions, schemas, microservices, or config flags naturally rationalizes their existence. A dedicated subagent has zero psychological or conversational investment in prior proposals.
- Context Contamination: Long-running conversation contexts contain discarded ideas, exploratory reasoning, edge-case anxieties, and premature compromises. This cognitive residue biases the evaluation toward maintaining accidental complexity. A fresh subagent receives only the target specification and the audit mandate, evaluating it with a clean slate.
- Preserving Impartiality & Objective Distance: Genuine red-teaming requires contrarian detachment. An isolated subagent is uninhibited by politeness or conversational continuity—it can objectively propose deleting 90% of a specification without defending earlier dialogue.
2. Default Delegation Policy & User Override
- DEFAULT Behaviour: When
uno-reverseis activated (via slash command//uno-reverse, skill loading, or prompt directive), the parent agent MUST spawn a dedicateduno-reversesubagent usinginvoke_subagentand immediately yield execution. - Inline Override: Subagent delegation is bypassed ONLY if the user explicitly requests an inline review (e.g., "review inline", "run uno-reverse in this session", or "do not use subagents"). In that case, the parent agent executes the 4-step workflow directly in the current context.
3. Delegation Pattern & Subagent Invocation
When delegating, select the subagent type:
research(Default): Ideal for read-only audits of design documents, PRDs, architectural RFCs, and codebases.self: Use when the audit also requires writing code, refactoring files, or generating alternative implementations directly on disk.Model: 'inherit': Preserve the parent session's reasoning effort and model capabilities.
Antigravity invoke_subagent Pattern:
{
"Subagents": [
{
"TypeName": "research",
"Role": "Uno-Reverse Auditor",
"Model": "inherit",
"Prompt": "You are an impartial, contrarian Uno-Reverse Auditor tasked with radical simplification and red-teaming.\n\n### Target of Audit:\n[FILE_PATH_OR_TARGET_DESCRIPTION]\n\n### Mandate & Mindset:\n- You have ZERO investment in prior discussions, earlier decisions, or existing abstractions.\n- Challenge every moving part, configuration flag, caching layer, and speculative interface.\n- Treat complexity as a liability: Reliability = 1 / (Moving Parts ^ 2).\n- Apply the 4 Inversion Principles (Subtraction Before Addition, Primitive Collapsing, Zero-Speculation YAGNI, Failure-Surface Inversion).\n- Respect Chesterton's Fence: understand the failure mode before eliminating essential safety.\n\n### Instructions:\n1. Execute the 4-Step Uno-Reverse Audit Workflow (Subtraction Test, Primitive Collapsing, Cache & State Probe, 10% MVP).\n2. Format your complete findings strictly as the 'Simplification Audit Scorecard':\n - Executive Inversion Summary (Proposed Complexity, Verdict, Code Reduction %)\n - The Cut List (Items to eliminate, rationale, what happens without it)\n - Collapsed Primitives (Disparate mechanisms -> Single primitive)\n - Minimalist Reference Design (10% MVP spec / code)\n - Chesterton Boundary Assessment (Essential complexity kept vs. pruned)\n3. Report your findings back using send_message."
}
]
}
Lifecycle & Non-Blocking Coordination:
- Dispatch & Yield: The parent agent invokes the subagent and immediately halts tool calls to end its turn. It never polls, sleeps, or busy-waits.
- Report Reception: When the subagent completes the audit and sends its report, the parent agent wakes up reactively.
- Synthesis & Presentation: The parent agent delivers the unvarnished scorecard to the user or uses the minimalist