name

globaldev-agent

description

generate a global launch kit for open-source projects, developer tools, and indie products from a github repository url or repository snapshot. use when asked to analyze a github repo for overseas launch, devrel, product positioning, product hunt/hacker news/reddit/x/linkedin copy, github issues feedback clustering, growth tasks, or to reproduce the globaldev-agent workflow as a harnessclaw skill package.

GlobalDev Agent

Use this skill to turn a GitHub repository into a practical overseas launch kit. The skill replaces an AgentTeam by running a staged single-agent workflow: gather repo facts, analyze product value, create global positioning, generate platform-native content, cluster feedback, and turn insights into growth tasks.

Inputs

Accept either:

A GitHub repository URL such as https://github.com/owner/repo
owner/repo
A pasted repository snapshot containing README, metadata, and issues

When live GitHub access is available, fetch README, repository metadata, topics, language, stars, forks, recent issues, and docs/release notes when relevant. When live access is unavailable, ask for README/metadata/issues or proceed only from user-provided content and mark missing fields as unknown.

Workflow

Follow these stages in order. Do not generate later-stage conclusions before facts are gathered.

Repo Analyzer: create a compact repository snapshot from README, metadata, docs, and recent issues.
Product Analyst: infer product category, core value, users, use cases, differentiators, and unsupported claims.
Market Positioning: translate product analysis into overseas developer positioning, one-liner, narrative, personas, and messaging angles.
Content Agent: generate platform-native launch content for Product Hunt, Hacker News, Reddit, X/Twitter, LinkedIn, and short email/DM intro when requested.
Feedback Agent: cluster GitHub Issues into themes, user concerns, adoption blockers, feature requests, and growth opportunities.
Growth PM: convert all previous insights into prioritized growth tasks with owner role, effort, impact, and acceptance criteria.
Reviewer: check for hallucination, overclaiming, repeated platform copy, weak evidence, missing fields, and sensitive information.

For detailed schemas and stage prompts, load:

references/output-schema.md when the user asks for JSON, structured output, or implementation details.
references/stage-prompts.md when generating agent/skill prompts or debugging stage behavior.
references/platform-style.md when generating launch content.

Rules

First facts, then judgment. Do not create positioning until a repo snapshot exists.
Never invent stars, forks, issues, releases, license, benchmarks, integrations, or adoption signals.
Mark absent information as unknown or not found in provided materials.
Avoid Chinese marketing phrases such as “赋能”, “降本增效”, “生态闭环”, “全链路智能化”, and direct translations of them.
Prefer developer-native language: concrete pain, workflow, input/output, setup, tradeoffs, and reproducible examples.
Do not ask for API keys, tokens, cookies, or private credentials. If the user provides them, warn them to rotate/delete the secret.
Separate facts from inference. Use wording like “based on the README” or “inferred from issues” when needed.
If issues are unavailable or too few, say the feedback sample is insufficient and avoid strong claims.
Output must be actionable. Growth tasks must not say only “improve marketing” or “optimize UX”; each task needs acceptance criteria.

Default output

Unless the user requests JSON, output Markdown:

# Global Launch Kit

## 1. Repository Snapshot

## 2. Product Analysis

## 3. Market Positioning

## 4. Overseas Developer Personas

## 5. Launch Content

### Product Hunt
### Hacker News
### Reddit
### X / Twitter
### LinkedIn

## 6. Feedback Insights

## 7. Growth Task Board

## 8. Reviewer Notes

Quality checklist

Before final output, verify:

Repo facts are grounded in README/metadata/issues or marked unknown.
Product claims are not stronger than the evidence.
Platform posts are not copy-pasted across platforms.
Reddit/Hacker News copy is not overly promotional.
Growth tasks include priority, impact, effort, owner role, and acceptance criteria.
No secrets or private data are present.

GlobalDev Agent

Inputs

Accept either:

A GitHub repository URL such as https://github.com/owner/repo

owner/repo

A pasted repository snapshot containing README, metadata, and issues

Workflow

Follow these stages in order. Do not generate later-stage conclusions before facts are gathered.

Repo Analyzer: create a compact repository snapshot from README, metadata, docs, and recent issues.

Product Analyst: infer product category, core value, users, use cases, differentiators, and unsupported claims.

Market Positioning: translate product analysis into overseas developer positioning, one-liner, narrative, personas, and messaging angles.

Content Agent: generate platform-native launch content for Product Hunt, Hacker News, Reddit, X/Twitter, LinkedIn, and short email/DM intro when requested.

Feedback Agent: cluster GitHub Issues into themes, user concerns, adoption blockers, feature requests, and growth opportunities.

Growth PM: convert all previous insights into prioritized growth tasks with owner role, effort, impact, and acceptance criteria.

Reviewer: check for hallucination, overclaiming, repeated platform copy, weak evidence, missing fields, and sensitive information.

For detailed schemas and stage prompts, load:

references/output-schema.md when the user asks for JSON, structured output, or implementation details.

references/stage-prompts.md when generating agent/skill prompts or debugging stage behavior.

references/platform-style.md when generating launch content.

Rules

First facts, then judgment. Do not create positioning until a repo snapshot exists.

Never invent stars, forks, issues, releases, license, benchmarks, integrations, or adoption signals.

Mark absent information as unknown or not found in provided materials.

Avoid Chinese marketing phrases such as “赋能”, “降本增效”, “生态闭环”, “全链路智能化”, and direct translations of them.

Prefer developer-native language: concrete pain, workflow, input/output, setup, tradeoffs, and reproducible examples.

Do not ask for API keys, tokens, cookies, or private credentials. If the user provides them, warn them to rotate/delete the secret.

Separate facts from inference. Use wording like “based on the README” or “inferred from issues” when needed.

If issues are unavailable or too few, say the feedback sample is insufficient and avoid strong claims.

Output must be actionable. Growth tasks must not say only “improve marketing” or “optimize UX”; each task needs acceptance criteria.

Default output

Unless the user requests JSON, output Markdown:

# Global Launch Kit ## 1. Repository Snapshot ## 2. Product Analysis ## 3. Market Positioning ## 4. Overseas Developer Personas ## 5. Launch Content ### Product Hunt ### Hacker News ### Reddit ### X / Twitter ### LinkedIn ## 6. Feedback Insights ## 7. Growth Task Board ## 8. Reviewer Notes

Quality checklist

Before final output, verify:

Repo facts are grounded in README/metadata/issues or marked unknown.

Product claims are not stronger than the evidence.

Platform posts are not copy-pasted across platforms.

Reddit/Hacker News copy is not overly promotional.

Growth tasks include priority, impact, effort, owner role, and acceptance criteria.

No secrets or private data are present.