ai-opportunity-scout

name	ai-opportunity-scout
description	Broad AI ecosystem scout. Use when an agent needs to map, rank, or analyze promising AI projects, applications, platforms, model/tool ecosystems, connectors, agent products, model-serving stacks, fine-tuning tools, eval/observability products, workflow builders, memory/context systems, and contribution opportunities across the full AI landscape.

AI Opportunity Scout

Treat the task as opportunity scouting, not grant hunting. The goal is to identify AI projects, products, and ecosystems that may become important and help the user quickly see where the interesting surface area is.

When the task depends on project momentum or ecosystem activity, prefer recent primary signals from GitHub issues, PRs, releases, docs, community channels, and company announcements instead of relying on memory.

For project-by-project diligence, read references/scouting-checklist.md.

Agent Compatibility

This skill is intentionally agent-agnostic.

• Use it from any capable coding or research agent, including Claude Code, OpenCode, OpenClaw, Codex, and similar tool-using assistants.
• If the host agent has web search or browser tools, use them for any fact that may have changed recently.
• If the host agent cannot browse, state that limitation clearly and avoid pretending that stale knowledge is current.
• Prefer short comparable summaries in broad scans so the user can choose what to drill into next.

Default Behavior

Unless the user clearly asks for a narrow slice, default to a broad scan first.

Broad scan means:

• Cover more of the AI landscape before going deep.
• Prefer breadth over diligence in the first pass.
• Include applications, agent products, frameworks, connectors, model tooling, fine-tuning stacks, data tooling, eval systems, memory/context systems, model infra, and workflow builders.
• Return 10-20 candidate projects when the space is broad enough.
• Keep summaries short so the user can choose what to drill into next.

Do not assume the user wants only open-source contribution targets. They may want:

• a landscape map
• applications to study
• integration layers to watch
• model and fine-tuning ecosystems
• upstream and downstream picks
• weird or interesting exploratory projects
• a shortlist to investigate later

Scope Framing

Before researching, identify which of these scopes fits best:

1. broad-ecosystem

   • Use when the user wants direction-finding, idea generation, or a wide scan.
   • Default output is a landscape map plus a long shortlist.

2. app-layer-products

   • Use for user-facing AI apps, agent products, copilots, workflow tools, knowledge tools, vertical AI products, browser agents, and local AI products.

3. infra-and-platform

   • Use for model gateways, inference stacks, training/fine-tuning tooling, evals, observability, deployment, data pipelines, provider connectors, and protocol layers.

4. context-memory

   • Use for agent memory, prompt compaction, conversation state, long-context tooling, knowledge recall, and context management systems.

5. contribution-opportunities

   • Use when the user explicitly wants OSS contribution wedges, maintainer-capacity gaps, or trust-building PR ideas.

6. theme-scout

   • Use when the user names a topic such as voice agents, computer use, coding agents, fine-tuning, open-source evals, or AI for legal.

If the user provides exclusions such as not too low-level, not databases, more product-side, or TypeScript-first, treat those as hard routing constraints.

Core Thesis

Use one of these theses depending on the scope:

• Broad ecosystem thesis:

  Find parts of the AI landscape where demand, novelty, or user behavior is changing quickly enough that new winners are still being formed.

• Contribution thesis:

  Find projects where ecosystem demand is growing faster than maintainer capacity.

• App-layer thesis:

  Find products where user pain is visible in the workflow, not just in the abstraction layer.

• Context-memory thesis:

  Find systems that treat context and memory as first-class product or agent behaviors, not just hidden implementation details.

Do not overfit on grants, fundraising, or hype. Prefer evidence of real usage, repeated demand, product friction, ecosystem pull, and visible room for a skilled outsider to add value.

Standard Workflow

1. Frame the search area.

   • First determine whether the user wants breadth or depth.
   • If they want breadth, scan multiple layers of the stack before selecting winners.
   • If they want depth, narrow to the named theme or product layer.
   • Record exclusions early and honor them throughout.

2. Build a landscape map.

   • Spread candidates across multiple layers when doing broad scans:
     • end-user applications
     • agent products
     • workflow builders
     • connectors and integration layers
     • context / memory systems
     • model tooling and fine-tuning stacks
     • observability and evals
     • inference / deployment platforms
     • weird or emergent experimental tools
   • If the user names an ecosystem such as OpenAI-compatible providers, LangChain, vLLM, MCP, Ollama, browser agents, coding agents, or memory systems, use it as one cluster, not the whole map.

3. Collect signals.

   • Funding and business signals are useful but not mandatory.
   • GitHub, issue, PR, release, Discord, docs, examples, and hiring signals are more important.
   • Prioritize recent signals when available.
   • In broad scans, fast pattern recognition matters more than full diligence.
   • In deep dives, inspect project-level details more carefully.

4. Classify each candidate.

   • What layer is it in?
   • Who is the user?
   • What workflow does it sit inside?
   • Is it infrastructure, product, model tooling, workflow, research-adjacent, or exploratory?
   • Is the interesting thing adoption, product design, ecosystem fit, contribution leverage, or technical novelty?

5. Score projects with the rubric below.

   • Use a lighter scoring pass for broad scans.
   • Use a full scoring pass for deep dives or contribution scouting.

6. Recommend next cuts.

   • In broad scans, tell the user which clusters deserve a second pass.
   • In contribution mode, recommend concrete wedges.
   • In theme mode, point to 3-5 must-open projects and why.

Opportunity Rubric

Score each project from 0 to 100.

Breakout Potential: 20 points

Look for signs that the project could become important even if it is not yet obviously monetized.

High-signal indicators:

• Rapid star, fork, contributor, or issue growth.
• Appears repeatedly in developer discussions, tutorials, benchmarks, or example repos.
• Solves a painful AI problem that is becoming more common.
• Sits near a fast-growing interface or product behavior such as agent tooling, memory, browser automation, evals, multimodality, fine-tuning, local AI, or workflow automation.
• Has credible founders, maintainers, users, or backers.

User or Ecosystem Relevance: 20 points

Look for projects that sit close to real workflows, visible pain, or important enabling layers.

High-signal indicators:

• Real users do work inside the product.
• Repeated references from builders, not just benchmark hobbyists.
• Clear fit in an upstream or downstream AI workflow.
• Strong adjacency to a growing category.

Contribution Leverage: 20 points

Look for places where a small external team can make visible impact.

High-signal indicators:

• Core maintainers are overloaded.
• Open issues are actionable but under-served.
• Docs, examples, adapters, integrations, tests, workflows, evals, deployment guides, or product flows are thin.
• PRs from external contributors get reviewed and merged.
• Maintainers respond constructively to well-scoped contributions.

Ecosystem Gap: 20 points

Look for missing connections that block adoption.

High-signal indicators:

• Requests for provider support, adapters, connectors, integrations, deployment recipes, workflow compatibility, memory visibility, or product features users repeatedly ask for.
• Incomplete support in major ecosystems or obvious upstream/downstream bridges.
• Repeated issues around streaming, tool calling, auth, rate limits, retries, observability, evals, context management, RAG behavior, UX friction, or deployment.

Timing: 15 points

Look for reasons this is a good moment to enter.

High-signal indicators:

• Recent funding, launch, pivot, benchmark, release, model update, or standardization wave.
• Recent complaints from users trying to adopt the project.
• New enterprise, cloud, or production deployment demand.
• Maintainer roadmap mentions ecosystem expansion.

Distinctiveness or Learning Value: 10 points

Look for projects that are worth studying even if they are not obvious contribution targets.

High-signal indicators:

• Novel product behavior or interaction design.
• Interesting memory, workflow, orchestration, or fine-tuning approach.
• A good reference implementation for an emerging theme.
• Reveals where the category is going.

Relationship Upside: 15 points

Look for natural paths to trust rather than immediate monetization.

High-signal indicators:

• Maintainers are accessible in issues, Discord, Slack, X, or community calls.
• Project has sponsors, enterprise users, or commercial company behind it.
• Contribution could put the team near users with production pain.
• The work could lead to reputation, consulting, employment, partnership, or future grant-like support without forcing the ask early.

Red Flags

Deprioritize projects when:

• Maintainers ignore external PRs.
• Issues are mostly hobbyist noise with little production demand.
• The architecture is unstable and contributions will be thrown away.
• The project has hype but no clear adoption path.
• The team is hostile to external contributors.
• The opportunity depends only on getting a grant.
• The project is only interesting because it is adjacent to AI, not because it solves an AI-native workflow.
• The user explicitly excluded this layer and the project only fits that excluded layer.

Search Queries and Signals

Broad-scan query clusters:

• site:github.com AI agent app open source GitHub
• site:github.com AI workflow builder open source GitHub
• site:github.com AI memory agent GitHub
• site:github.com LLM fine tuning open source GitHub
• site:github.com local AI app open source GitHub
• site:github.com AI observability open source GitHub
• site:github.com browser agent open source GitHub
• site:github.com coding agent open source GitHub
• site:github.com AI connector integration open source GitHub

Contribution-mode query clusters:

• site:github.com AI agent framework open issues adapter support
• site:github.com litellm provider support tool calling streaming issue
• site:github.com mcp server open issues integration
• "OpenAI-compatible" "does not work" github issue
• "tool calling" "streaming" "provider" github issue
• "any plan to support" "provider" "AI"
• "helm chart" "LLM" github issue
• "enterprise deployment" "AI" github issue
• "MCP" "integration" "not supported"

Context-memory query clusters:

• "agent memory" github open source
• "context window" agent github issue
• "conversation history" AI app github issue
• "memory" "workspace" "agent" github
• "prompt compaction" github agent
• "long context" "AI" github issue

App-layer query clusters:

• "AI app" "self-hosted" github
• "knowledge base" "AI agent" github
• "workflow builder" "AI" github
• "browser agent" github issue
• "coding agent" github release
• "AI assistant" "multi-user" github

When scanning GitHub, inspect:

• Recent issues, not just total issue count.
• Oldest unclosed integration issues.
• Maintainer response time.
• External PR merge rate.
• Contributor concentration.
• Docs and examples quality.
• Release cadence.
• Whether users describe production blockers.

Output Shapes

Prefer one of these outputs based on the user request.

1. Broad Landscape Map

Use for wide scans. Group projects by layer and keep each description to one or two lines.

Suggested structure:

## Landscape Map

### App-Layer Products
- Project: why it is worth looking at

### Agent and Workflow Tools
- Project: why it is worth looking at

### Context and Memory
- Project: why it is worth looking at

### Model / Fine-Tuning / Serving
- Project: why it is worth looking at

### Interesting Experimental Tools
- Project: why it is worth looking at

## Best Next Cuts
1. Cluster to drill into next
2. Cluster to drill into next
3. Cluster to drill into next

2. Ranked Projects

Use for deeper comparison or contribution scouting.

## Top Opportunities

### 1. <project>
**Thesis:** <why this could matter>
**Current signal:** <growth, demand, release, ecosystem relevance>
**Gap:** <specific missing integration/docs/deployment/support>
**Best first move:** <small realistic contribution or reason to study>
**Bigger wedge:** <medium contribution, product angle, or research angle>
**Why it is worth studying:** <what this teaches you about the category>
**Opportunity score:** <0-100>
**Risk:** <main reason this may not work>

3. Artifact Output

Use when the user wants the results as a document or page.

• HTML page
• Markdown memo
• table or backlog

For broad scans, favor a long shortlist over over-arguing the top three.

Tone and Strategy

Be direct and commercially aware, but do not sound extractive. Frame the team as useful contributors or careful researchers who are trying to understand where real value is being created.

Avoid language that assumes immediate grants, bounty money, or guaranteed consulting. Prefer phrases like:

• natural upside
• relationship path
• trust-building contribution
• contribution wedge
• ecosystem leverage
• production adoption blocker

BD Guidance

Only recommend outreach after there is a credible contribution or concrete diagnosis.

A good sequence is:

1. Comment helpfully on an issue.
2. Submit a small PR.
3. Ask maintainers whether a related gap would be useful.
4. Ship a medium-sized contribution.
5. Only then mention that the team can help maintain this area or support users adopting it.

Do not lead with we want a grant.