# Idea Validation Agent

> Validate article links, social posts, or Obsidian ideas as side-project or content-product opportunities. Use when the user asks Codex to read an article/link, match it against the Obsidian knowledge base or ideas tracker, search X/Twitter, Xiaohongshu, Reddit, V2EX, GitHub, Exa, or other external sources for examples and risks, and produce a practical MVP, monetization, or feasibility plan. Also use for daily/weekly automation that reads ideas and designs evidence-backed implementation plans.

- Skill: `anklecrusher/idea-validation-agent` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add anklecrusher/idea-validation-agent`
- Raw SKILL.md: https://api.skillmd.com/api/skills/anklecrusher/idea-validation-agent/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: anklecrusher (https://skillmd.com/u/anklecrusher)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/anklecrusher/idea-validation-agent

---


# Idea Validation Agent

Turn an article, post, or `ideas/` entry into an evidence-backed project plan.

## Core Rule

Do not treat social heat as proof. Assign each source a role:

- **Input article/post**: extract the claim, audience, pain, solution, and commercialization angle.
- **Obsidian**: check fit with the user's existing knowledge, ideas, methods, practices, and questions.
- **X / Twitter**: find frontier examples, builders, tools, pricing, and overseas implementation patterns.
- **Xiaohongshu**: validate Chinese demand, content formats, comments, service intent, and monetization clues.
- **Reddit / technical communities**: inspect implementation risks, user complaints, failure modes, alternatives, and hidden costs.
- **Other sources**: use Exa, GitHub, official docs, V2EX, Stack Overflow, product forums, Hacker News, YouTube/Bilibili, or RSS when they fit the question better.

## Workflow

### 1. Parse the User Input

Identify the task type:

- Article/link validation
- Existing `ideas/tracker` item validation
- Batch daily/weekly idea review
- Follow-up on a previous validation report

Extract or ask only if necessary:

- Link or idea ID
- Target user group
- Desired market: Chinese, global, or both
- Expected deliverable: service, template, tool, course, content account, community, consulting, or unclear

### 2. Read the Source

Pick the reading route:

- X / Twitter post: use `twitter tweet URL_OR_ID`.
- X Article: use `twitter article URL_OR_ID`.
- Generic webpage: use Jina Reader or web-reader.
- WeChat article: prefer Exa crawling/search.
- GitHub or docs: use GitHub, Exa, or official docs.

Extract:

- Claim
- Target user
- Pain point
- Proposed solution
- Commercialization angle
- Assumptions that require evidence
- Search keywords in Chinese and English

### 3. Match Obsidian

Use Obsidian MCP when available. Read, in this order:

1. `wiki/index.md`
2. `wiki/overview.md`
3. `schema/idea-tracking.md` if writing or reading `ideas/`
4. `ideas/tracker.md` and current weekly review when the user wants idea continuity
5. Directly relevant `methods/`, `practice/`, `questions/`, `concepts/`, or `entities/`

Check:

- Existing related ideas
- Reusable methods or practices
- Known risks or decision questions
- Whether this belongs in `ideas/`, `practice/`, `methods/`, or only the conversation

Respect write boundaries:

- Do not write Obsidian unless the user asks.
- For idea automation, default writes only to `ideas/`, following `schema/idea-tracking.md`.
- Do not write secrets, cookies, tokens, or account identifiers beyond high-level status.

### 4. Search External Evidence

Use Agent-Reach routes where installed. Prefer structured output such as `--yaml` or `--json`.

X / Twitter:

```bash
twitter search "query" -n 10 --yaml
twitter user-posts username -n 20 --yaml
twitter tweet URL_OR_ID --yaml
twitter article URL_OR_ID --yaml
```

If your Agent-Reach install exposes Windows wrapper scripts instead of PATH commands, replace `twitter` with the full wrapper path, for example:

```powershell
& "$env:USERPROFILE\.agent-reach\bin\twitter.cmd" search "query" -n 10 --yaml
```

If X search fails because upstream GraphQL changed, use feed/user-posts, web search, or Exa.

Xiaohongshu:

```bash
xhs search "query" --page 1 --yaml
xhs read NOTE_ID_OR_URL --yaml
xhs comments NOTE_ID_OR_URL --yaml
```

Use search results or full note URLs for reads; do not invent bare note IDs. If the command needs to write `search_sessions.json`, run it in an environment where the agent has permission to write its local cache.

Reddit:

```bash
rdt search "query" --limit 10 --yaml
rdt read POST_ID --yaml
rdt sub subreddit --limit 20 --yaml
```

Reddit does not need login for basic search/read. Use it when the idea has technical, platform, workflow, or user-complaint risk.

Other sources:

- Use Exa for English/technical web and code evidence.
- Use GitHub search/issues for implementation friction.
- Use V2EX for Chinese technical user sentiment.
- Use official docs for API, policy, pricing, and platform rules.

### 5. Synthesize the Plan

Produce a decision, not a pile of links. Use `references/report-template.md` for the report structure.

Decision labels:

- **Continue**: enough demand and feasible MVP path.
- **Observe**: promising but key evidence missing.
- **Pause**: weak demand, unclear buyer, or high execution risk.

Always include:

- Evidence table by platform/source
- Pain point synthesis
- MVP recommendation
- 3-day action plan
- Risk and mitigation
- Obsidian update recommendation

## Automation Mode

For daily/weekly automation:

1. Read `schema/idea-tracking.md`.
2. Read `ideas/tracker.md` and the relevant weekly review.
3. Select only `待验证` or `可行动` ideas.
4. Run a shallow evidence pass for each idea, or a deep pass for the top 1-3 ideas.
5. Write only to `ideas/` if the user or automation explicitly allows writes.
6. Suggest, but do not automatically create, `wiki/practice`, `wiki/methods`, or `wiki/questions` pages.

## Quality Bar

- Cite source URLs or local note paths in the answer.
- Distinguish source evidence from inference.
- Prefer recent, primary, or platform-native evidence.
- Do not over-index on X excitement or Xiaohongshu vanity metrics.
- Identify the buyer, not only the user.
- Avoid recommending a build before the first acquisition path is clear.
- If evidence is weak, say so and specify the next search that would reduce uncertainty.

## Report Template

Read `references/report-template.md` when producing the final validation report.