comprehensive-search

Comprehensive Search

English · 中文

A coverage-oriented research skill for AI agents that maps the search space, expands queries adaptively, seeks counter-evidence by default, tracks source independence, audits coverage gaps, and reports what remains unknown — before drawing conclusions.

Core capabilities

• 🗺️ Search-space mapping — breaks a question into information dimensions instead of firing random queries.
• ⚔️ Adversarial search by default — actively hunts counter-evidence, failure cases, and biases, not just supporting material.
• 🔗 Source-independence tracking — distinguishes "two web pages" from "two truly independent evidence chains."
• 📊 Coverage audit & gaps — shows what is supported, what is partial, what is unresolved, and what was simply not found.
• 🛑 Operational-saturation stopping rule — stops on low information gain instead of looping forever or pretending the web is exhausted.

Quick start

1. Drop the comprehensive-search/ folder into your agent's skills/ directory (see Installation).
2. Ask the agent something like "Comprehensively research whether X is worth buying" or "Map the information space around Y before I decide."
3. The agent returns a structured report: search space → coverage matrix → counter-evidence → gaps → confidence, plus a Saturation status.

What it is not. Comprehensive Search does not claim to exhaust the entire internet. It aims for systematic coverage under available sources, tools, time, and access constraints. Search engines have incomplete indexes, paywalls, closed communities, deleted pages, geo/language limits, and tool restrictions. The skill is honest about those blind spots.

Why

Ordinary AI search tends to follow a failure pattern:

find a few results
   → form a conclusion quickly
   → keep searching around that conclusion
   → repeatedly cite the same source
   → ignore counter-examples
   → mistake "not found" for "does not exist"

Comprehensive Search is built to break that pattern. It treats coverage before conclusion as a discipline, not a slogan.

Before / After

A first-screen demo of the method and output shape. This is not a real market study — numbers are illustrative placeholders, not research findings.

Question: "Is RWA (Real-World Asset tokenization) worth a long-term allocation?"

Ordinary agent output

RWA market growing fast 🚀
ONDO leading the sector
Institutions adopting tokenization
Regulatory clarity improving
→ Strong buy / strong long-term thesis

(Four bullish bullet points, no counter-evidence, no source accounting, no unknowns.)

Comprehensive Search output

Search Space
├─ Market size & growth
├─ Definition differences (what counts as "RWA"?)
├─ Real on-chain adoption (not just announcements)
├─ Revenue / token value capture
├─ Regulation (per jurisdiction)
├─ Failure cases / stalled projects
├─ Competing approaches
├─ Counter-arguments
└─ Valuation

Coverage
✅ 7 sufficiently supported
🔁 2 independently corroborated
⚠️ 2 partially supported (single source)
❓ 1 unresolved (conflicting adoption figures)

Key gaps
- On-chain fee/revenue attribution not standardized → ⚠️ partial
- Some "adoption" claims are pilots, not production → ❓ unresolved
- Non-English regulatory sources not fully covered → language blind spot

Counter-evidence
- Multiple announced partnerships remained pilots / wound down
- Token value capture often decoupled from protocol revenue
- Custody, compliance, and oracle risk under-documented

Saturation status: Reached (within English/web + public sources);
  blind spots: paywalled databases, non-English regs, deleted pages.

Same question, completely different rigor.

How it works

Question
   ↓
Research Contract      (goal, scope, risk level, fact vs. judgment)
   ↓
Search Space          (map dimensions; do NOT blindly apply all)
   ↓
Adaptive Search        (enumerate space, pick queries by information gain)
   ↓
Evidence + Counter-evidence   (adversarial search is mandatory)
   ↓
Coverage Audit        (supported? single-source? conflict? not searched?)
   ↓
Operational Saturation        (stop on low gain — heuristic, not proof)
   ↓
Synthesis + Unknowns  (state what we do NOT know)

Installation

Comprehensive Search follows the standard Agent Skills layout: a folder named comprehensive-search/ containing a SKILL.md, optional references/, examples/, and a LICENSE.

Install with your AI agent

If your AI agent can access the web and manage local files, you can ask it to install the skill automatically.

Copy this prompt (English):

Open https://github.com/ziyue085/comprehensive-search, read the repository, install `comprehensive-search` into the appropriate Skills directory for your current agent environment following the Agent Skills specification, then verify that `SKILL.md` is successfully recognized and report the result.

Or this prompt (中文):

请打开 https://github.com/ziyue085/comprehensive-search ，读取仓库内容，并按照 Agent Skills 规范将 `comprehensive-search` 自动安装到你当前环境可用的 Skills 目录；安装完成后验证 `SKILL.md` 已被正确识别，并报告安装结果。

The agent must have web access and permission to write to the local Skills directory. Installation paths vary by agent platform, so this repository intentionally does not hard-code a vendor-specific path.

Generic Agent Skills installation

Copy or clone the folder so your agent's skill loader can discover it:

<skills-dir>/comprehensive-search/
├── SKILL.md
├── README.md
├── LICENSE
├── references/
└── examples/

Then restart / re-index the agent so it picks up the new name + description.

Manual installation

Download the folder and place it under your platform's skills directory (e.g. a user-level or project-level skills/ path). No build step, no dependencies.

Compatibility

This skill uses only natural-language instructions and the references/ progressive-disclosure pattern defined by the Agent Skills spec. It works with any agent that supports the Skill mechanism and has web/search access. Additional domain-specific sources (financial data, academic search, registries) improve coverage but are not required.

No specific vendor/platform is claimed as "officially supported" here. If your agent loads SKILL.md and can search, it can use this skill.

Usage

Trigger phrases / requests:

• "Comprehensively research whether [X] is worth [buying / adopting / trusting]."
• "Do a comprehensive search before answering — don't miss counter-evidence."
• "Map the information space around [topic]; show me gaps and unknowns."
• "Search both supporting and opposing evidence for [claim]."
• "Before I decide on [decision], make sure we're not missing failure cases or biases."
• "Audit our coverage: what's single-source, what's conflicting, what's unresolved?"

Not for: quick single-fact lookups, trivia, or confirming an answer the user already has.

Limitations

• Not a completeness guarantee. It does not prove the entire information universe was exhausted.
• Tool-bound. Coverage depends on the agent's search tools and their index quality.
• Access-bound. Paywalled, private, deleted, or closed-community sources create blind spots.
• Independence is sometimes inferred. When provenance is unclear, source independence can only be judged heuristically.
• Operational saturation ≠ universal completeness. It is a stopping heuristic under current constraints.
• High-stakes domains still need humans. For legal, medical, or financial decisions, use professionals and authoritative databases; this skill organizes and audits evidence, it does not replace expert judgment.

Philosophy / Design principles

Coverage before conclusion
Evidence independence over source count
Counter-evidence by default
Unknowns are part of the answer
Search depth should follow information gain

Project structure

comprehensive-search/
├── SKILL.md                  # Entry point: 9-phase workflow + stopping rule
├── README.md                # This file (English)
├── README.zh.md             # Chinese translation
├── LICENSE                  # MIT
├── CHANGELOG.md             # Release notes
├── references/
│   ├── dimensions.md        # Search-space dimensions + adaptive query rules
│   ├── source-quality.md    # Source tiers, independence, primary-source rule
│   ├── evidence-status.md   # 5-state evidence model + report template
│   └── saturation.md        # Operational saturation + stopping details
└── examples/
    ├── example-1-investment.md
    └── example-2-tech-comparison.md

License

Released under the MIT License. Copyright © 2026 Comprehensive Search contributors.