菜单
Dispatches research agents to find real projects, papers, and documented failures before proposing any feature, architecture, or technology choice. Triggers on creative/design tasks, not on implementation of already-designed specs.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Validates completed work against defined acceptance criteria. Use after completing a task that has specific success criteria defined in issues, specs, or task descriptions.
Generates edge case, failure mode, and spec-driven test cases. Covers boundary values, nil inputs, concurrency, resource exhaustion, malformed data, and requirement-linked traceability tests.
Detects and prevents shortcut rationalization — when the agent is about to skip a step, simplify a requirement, or optimize away a constraint. Triggers proactively when rationalization patterns are noticed.
Provides a persistent cross-session task queue for work that should be done but isn't blocking current tasks. Enables background and future-session work tracking.
Safely integrates commits from parallel agent branches using sequential cherry-pick. Use after parallel work completes in isolated branches or worktrees.
Audits tests for concurrency safety — race conditions, shared mock state, cleanup ordering. Use when writing tests that involve goroutines, async operations, or shared mutable state.
| name | vibe-start-informed |
| description | Dispatches research agents to find real projects, papers, and documented failures before proposing any feature, architecture, or technology choice. Triggers on creative/design tasks, not on implementation of already-designed specs. |
| user-invocable | true |
Other people's documented failures count as observed failure. Research SOTA before proposing anything.
Not for: Implementing already-designed specs, bug fixes, or tasks with clear requirements.
BEFORE proposing any approach, dispatch a research agent to find real papers, real projects, and real production systems that solve the same (or adjacent) problem.
digraph start_informed {
"Design question identified" [shape=doublecircle];
"Decompose into sub-problems" [shape=box];
"Dispatch research agent (WebSearch)" [shape=box];
"For each sub-problem find:" [shape=box];
"Synthesize imported failures" [shape=box];
"NOW propose approaches" [shape=doublecircle];
"Design question identified" -> "Decompose into sub-problems";
"Decompose into sub-problems" -> "Dispatch research agent (WebSearch)";
"Dispatch research agent (WebSearch)" -> "For each sub-problem find:";
"For each sub-problem find:" -> "Synthesize imported failures";
"Synthesize imported failures" -> "NOW propose approaches";
}
For each sub-problem, find:
| Requirement | Minimum | Example |
|---|---|---|
| Named projects | 2+ with stars/adoption | "Graphiti (23k stars)" |
| Papers with venue | 1+ with year | "RAPTOR (ICLR 2024)" |
| Mechanism described | What it actually does | Not marketing copy |
| Known failures | What went wrong | "Flat RAG failed at multi-hop" |
| Applicability assessment | How it maps to our system | Specific, not generic |
| Thought | Reality |
|---|---|
| "I already know the best approach" | You know ONE approach. Research finds others. |
| "This is too niche for research" | If someone built it, someone documented failures. |
| "Research will slow us down" | Uninformed design causes rewrites. Research prevents them. |
| "The textbook answer is obvious" | JK: "These are very very weak suggestions." Do real research. |
| "Let me just propose something" | Proposals without research are guesses with confidence. |
Alien Chang-Unni Constitution v0.1, Law 5 (amended in Session 2):
Start Informed — leverage SOTA research as collective observed failure. Simplify only when observed complexity causes problems in practice. Other people's documented failures count as observed failure.
The amendment was triggered when a constitutional audit revealed every design decision violated the original "Start Stupid" law. The law was changed because importing proven approaches isn't premature complexity — it's standing on shoulders.