Tech Trend Radar

Use when generating current technology, AI model ranking, product, startup, funding, regulation, or market trend reports from recent industry updates

Security Scan

The skill is instruction-only and its requested behaviors (web research, model/ranking checks, GitHub momentum, and producing time-windowed reports) align with its description; it does not request credentials, install code, or access unexpected system resources.

Purpose & Capability

Name/description ask for up-to-date trend reports and SKILL.md gives a research workflow and source priorities that match that purpose. There are no unrelated environment variables, binaries, or installs demanded.

Instruction Scope

Instructions require re-running live searches, checking leaderboards, GitHub Trending, and Anthropic/OpenAI announcements — i.e., network/browsing access is expected. The skill also suggests saving reports under memory/research/..., which is consistent with producing outputs but means reports may be written to the agent's memory/storage.

Install Mechanism

No install spec and no code files — instruction-only; nothing is downloaded or written to disk by the skill itself.

Credentials

The skill declares no required environment variables, credentials, or config paths. The workflow references public sources but does not request secrets or unrelated credentials.

Persistence & Privilege

always:false and default autonomous invocation are normal. The only persistence hint is the suggested report path (memory/research/...), meaning the agent may store generated reports in its memory — this is expected but be aware of stored report content.

Assessment

This skill is an instruction-only research template that asks the agent to fetch and synthesize public, recent signals (leaderboards, GitHub, company blogs). Before installing, confirm you are comfortable with the agent having network/browsing access and storing generated reports in its memory/storage. If you do not want live web access, do not enable browsing or deny the agent the HTTP/browse tools. There are no requested secrets or installs, so the main privacy consideration is what the agent will save into its memory and what external sites it will visit during research.

Trend Radar

Generate an industry trend report from recent, credible technology and market updates. Prioritize fresh signals, source quality, model reranking, GitHub momentum, business impact, and clear next actions.

When to Use

Use this skill when the user asks for:

• Weekly, monthly, or quarterly technology trends
• AI, SaaS, developer tools, cloud, security, startup, or regulation updates
• Model reranking, leaderboard changes, and current top model summaries
• Claude managed agents, Claude agent teams, and Anthropic agent platform updates
• OpenAI, Anthropic, Google, Microsoft, Meta, xAI, Alibaba Qwen, Zhipu GLM, and major AI lab strategy updates
• GitHub Trending and open-source momentum summaries
• Market signal summaries for planning, investing, product strategy, or competitive tracking
• A concise report that distinguishes major shifts from ordinary news

Trigger examples:

Write this week's AI trend report
Show current top 10 AI models and ranking changes
Track Claude managed agents and agent teams updates
Generate a trend radar for developer tools
Summarize GitHub Trending for AI agents this month
Summarize recent market trends in AI agents
What changed in SaaS and AI this week?

Research Workflow

1. Define the report period. Default to 1w unless the user specifies 1w, 1m, 3m, or an exact date range.
2. Search by category, not only by generic keywords. Combine company names, product names, ecosystem terms, and event types.
3. Check model leaderboards for reranking, new entrants, score deltas, and category shifts.
4. Check Anthropic and Claude agent platform updates, especially managed agents, agent teams, enterprise controls, connectors, and orchestration features.
5. Check GitHub Trending and repository momentum for relevant categories, languages, and keywords.
6. Prefer primary and high-signal sources before commentary.
7. Cross-check major claims with at least two credible sources when possible.
8. Deduplicate repeated coverage of the same announcement.
9. Score each item by novelty, impact, source quality, ranking movement, open-source traction, and actionability.
10. Separate confirmed facts from interpretation. Label uncertain items as Watch.
11. Re-check the latest available sources every run, even if the same topic appeared in a previous report.

Time Windows

Use three standard windows to separate short-term noise from durable trends:

Window	Use For	Signal Focus
1w	Weekly scan	launches, releases, incidents, funding, GitHub Trending spikes
1m	Monthly pattern	repeated adoption, repo growth, new categories, competitor movement
3m	Quarterly shift	platform changes, durable market direction, ecosystem consolidation

When reporting multiple windows, compare them explicitly: 1w = new spike, 1m = traction, 3m = durable direction.

Freshness and Continuity Rules

Each report must include the newest available information for its period. Do not omit an important topic because it appeared in an earlier report.

• Always refresh: rerun searches for model rankings, company strategy, product launches, regulatory actions, GitHub Trending, and agent platform updates.
• Show the delta: when a topic repeats, state what changed since the previous mention: new date, new source, rank movement, adoption signal, pricing change, policy update, or ecosystem reaction.
• Label continuity: mark repeated topics as Continuing, Updated, or Resolved.
• Prefer dated evidence: include publication dates or checked timestamps for all current claims.
• Reject stale summaries: do not rely only on an older saved report when the user asks for the latest period.

Required Depth

Reports should be concise but not shallow. For monthly or quarterly reports, each selected signal must include enough detail to support judgment:

• What happened: 2-4 concrete facts with dates, actors, product names, or metrics
• Why it matters: strategic, technical, market, regulatory, or developer impact
• Evidence: at least one direct source link; use two sources for high-impact claims when possible
• Core change: what is different from the previous period or previous report
• Action: monitor, test, migrate, compare, budget, update policy, or ignore

Avoid top-list-only reports. A Top 10 table is a navigation aid, not the analysis.

Source Priority

Use higher-priority sources first:

Priority	Source Type	Examples
1	Official announcements	company blogs, product release notes, API changelogs, pricing pages
1	Primary technical artifacts	GitHub releases, RFCs, benchmark reports, model cards, research papers
1	Anthropic / Claude updates	Anthropic news, Claude docs, Claude Code docs, API changelog, enterprise admin docs
1	Model leaderboards	LMArena/Text Arena, Artificial Analysis, LiveBench, SWE-bench, OpenRouter rankings
1	GitHub momentum	GitHub Trending, stars, forks, releases, issue velocity, contributor activity
2	Business records	funding announcements, acquisition news, earnings calls, regulatory filings
2	Trusted industry media	The Verge, TechCrunch, The Information, Stratechery, VentureBeat, The Pragmatic Engineer
3	Community signals	Hacker News, Reddit, X posts, Discord/Slack communities, GitHub stars/issues

Community signals should support a trend, not be the only evidence for a high-impact claim.

Anthropic / Claude Agent Signals

Track Claude agent platform changes as a dedicated signal family. Include both official product updates and ecosystem reactions when they affect how teams build, deploy, or govern agents.

Monitor these topics:

• Claude managed agents: hosted agents, managed execution, background tasks, schedules, permissions, observability, and lifecycle management
• Claude agent teams: multi-agent collaboration, team workspaces, handoffs, shared memory, role assignment, approval flows, and delegation patterns
• Claude Code and developer agents: CLI updates, IDE integrations, repo context, code review, test execution, MCP support, and automation workflows
• Enterprise controls: admin policy, audit logs, identity, data retention, connectors, compliance, sandboxing, and human-in-the-loop review
• API and platform primitives: tool use, computer use, MCP, files, sessions, workflows, web search, batch jobs, and rate/pricing changes

For each Claude agent signal, capture:

• What changed: product, API, docs, pricing, availability, or policy
• Affected users: individual developers, product teams, enterprise admins, platform teams, or agent builders
• Core change: managed execution, multi-agent coordination, governance, connector expansion, or deployment workflow
• Evidence: official Anthropic source first; add credible secondary analysis only when useful
• Action: try, migrate, monitor, update policy, or compare with OpenAI/Codex/Gemini/Copilot alternatives

Company Strategy Signals

Track strategic changes from major AI companies every run, especially OpenAI, Anthropic, Alibaba Qwen, and Zhipu AI. Include Google, Microsoft/GitHub, Meta, xAI, Mistral, DeepSeek, Moonshot AI, Baidu, Tencent, ByteDance, and other major labs when relevant.

For each company strategy item, capture:

• Company: OpenAI, Anthropic, Alibaba/Tongyi Qianwen/Qwen, Zhipu AI/GLM, Google Gemini/Gemma, Microsoft/GitHub, Meta, xAI, Mistral, DeepSeek, Moonshot AI, Baidu, Tencent, ByteDance, or other
• Latest move: product launch, model release, pricing, partnership, safety policy, enterprise push, developer platform shift, acquisition, hiring, or regulation response
• Strategic direction: models, agents, consumer apps, enterprise, coding, infrastructure, safety, open source, or distribution
• Evidence: official source first; add credible analysis or market data when useful
• Implication: what builders, buyers, investors, or competitors should do next

Monthly reports must include a Company Strategy Watch section with at least OpenAI, Anthropic, Alibaba Qwen, and Zhipu AI, even if there is no major change. In that case, write No major new public move found and cite the latest checked sources.

Model Reranking Signals

Track AI model ranking movement as a dedicated signal. Always state which leaderboard is used because rankings measure different things:

Leaderboard	Measures	Use For
LMArena / Text Arena	Human preference battles	General chat quality and user preference
Artificial Analysis	Composite intelligence, speed, cost	Balanced capability and deployment comparison
LiveBench	Dynamic benchmark performance	Reasoning, coding, math, and contamination-resistant evals
SWE-bench	Real-world coding tasks	Coding agent and software engineering capability
OpenRouter Rankings	Real API usage	Developer adoption and production demand
Open-weight model leaderboards	Open model performance	Gemma, Llama, Qwen, Mistral, DeepSeek, GLM comparison
Chinese model leaderboards	Domestic model performance	Qwen, GLM, DeepSeek, Kimi, ERNIE, Hunyuan, Doubao comparison

For each reranking item, capture:

• Current Top 10: rank, model, provider, score or usage metric, source, timestamp
• Core change: new entrant, rank gain/loss, score delta, usage spike, category win, or pricing-driven adoption
• Why it changed: model release, benchmark update, price change, context window, tool use, coding quality, availability
• Impact: product choice, API routing, coding workflow, cost strategy, or competitive positioning
• Confidence: high when leaderboard source and timestamp are clear; lower when based on secondary summaries

Monthly and quarterly reports must include a full Top 10 model table, not only a short summary. Include rank, model, provider, metric, leaderboard, checked timestamp, rank change if available, and the reason for notable movement.

Always check whether Chinese models appear in current rankings or category leaderboards. Track at least Qwen/通义千问, GLM/智谱, DeepSeek, Kimi/Moonshot, ERNIE/文心, Hunyuan/混元, and Doubao/豆包 when they are relevant.

GitHub Trending Signals

Track GitHub Trending as an early indicator, especially for developer tools, AI agents, infrastructure, and security. Search by relevant languages and topics such as Python, TypeScript, Rust, Go, AI, LLM, agent, RAG, MCP, security, and devtools.

For each notable repository, capture:

• Repository: owner/name and link
• Category: AI/ML, Developer Tools, Cloud/Infra, Security, or other
• Window: 1w, 1m, or 3m
• Momentum: stars gained, forks, release activity, contributors, issue velocity
• Why it matters: product category, technical novelty, adoption signal, or ecosystem relevance
• Caveat: hype, demo-only status, weak maintenance, unclear license, or duplicated project

Monitoring Categories

AI/ML

• Models: OpenAI, Anthropic, Google Gemini, Google Gemma, Meta Llama, Mistral, xAI, DeepSeek, Alibaba Qwen/通义千问, Zhipu GLM/智谱, Kimi/Moonshot, Baidu ERNIE/文心, Tencent Hunyuan/混元, ByteDance Doubao/豆包
• Agent frameworks: LangChain, LangGraph, AutoGen, CrewAI, OpenAI Agents SDK
• Capabilities: multimodal AI, coding agents, RAG, long context, reasoning, evaluation
• Claude agents: managed agents, agent teams, Claude Code, MCP, tool use, computer use, enterprise controls
• Business signals: AI startup funding, enterprise adoption, AI pricing, AI regulation
• Company strategy: OpenAI, Anthropic, Alibaba/Tongyi Qianwen/Qwen, Zhipu AI/GLM, Google Gemini/Gemma, Microsoft/GitHub, Meta, xAI, Mistral, DeepSeek, Moonshot AI, Baidu, Tencent, ByteDance
• Benchmarks and reranking: LMArena/Text Arena, Artificial Analysis, LiveBench, SWE-bench, OpenRouter, model cards, eval reports

Developer Tools

• Coding assistants: Codex, GitHub Copilot, Cursor, Claude Code, Windsurf
• Dev platforms: GitHub, GitLab, Linear, Vercel, Netlify, Supabase
• Tooling: CI/CD, observability, testing, package managers, IDE extensions
• Agent platforms: Claude managed agents, Claude agent teams, Codex agents, Copilot coding agent, Gemini agents
• GitHub Trending: fast-growing repos, new frameworks, agent tools, MCP servers, CLIs, infra templates

SaaS/Productivity

• Notion, Slack, Linear, Atlassian, Salesforce, HubSpot, Microsoft 365, Google Workspace
• AI features, workflow automation, collaboration, pricing, enterprise adoption

Cloud/Infrastructure

• AWS, Azure, Google Cloud, Cloudflare, Docker, Kubernetes, serverless, databases
• GPU supply, inference platforms, edge compute, cost optimization

Security/Privacy

• Vulnerabilities, supply-chain attacks, identity, compliance, AI safety, data governance

Startup/Funding

• Seed to IPO funding, acquisitions, shutdowns, category creation, Korean startups, global AI startups

Agent Platform Ecosystem

• Anthropic/Claude, OpenAI/Codex, Alibaba Qwen/通义千问, Zhipu GLM/智谱, Google/Gemini/Gemma, Microsoft/GitHub Copilot, MCP ecosystem, managed agents, agent teams, community showcases, OpenClaw, HarnessClaw

Scoring Criteria

Score	Meaning	Criteria
9-10	Critical	Industry-wide shift, immediate action needed, major platform or regulation change
7-8	High	Major company move, strong adoption signal, pricing or capability change
5-6	Medium	Useful signal, early traction, worth monitoring
1-4	Low	Minor update, rumor, narrow relevance

Raise the score when a trend affects roadmap, budget, hiring, compliance, open-source adoption, or competitive positioning. Lower it when evidence is weak, duplicated, only speculative, or based only on short-lived GitHub hype.

Report Format

Use this structure for readable output:

# Trend Radar YYYY-WXX

**Period**: YYYY-MM-DD to YYYY-MM-DD  
**Window**: 1w / 1m / 3m  
**Generated**: YYYY-MM-DD  
**Scope**: AI/ML, Developer Tools, SaaS, Cloud/Infra, Security, Startup/Funding, Agent Platform Ecosystem  
**Signal Count**: 00 reviewed, 00 selected

## Executive Summary

- **1w signal**: ...
- **1m signal**: ...
- **3m signal**: ...
- **Model reranking**: ...
- **Claude agent signal**: ...
- **Company strategy**: ...
- **Updated from previous report**: ...
- **Biggest opportunity**: ...
- **Biggest risk**: ...
- **Recommended action**: ...

## Top Signals

| Rank | Signal | Category | Window | Score | Status | Why It Matters |
|------|--------|----------|--------|-------|--------|----------------|
| 1 | ... | AI/ML | 1w | 9 | Confirmed | ... |
| 2 | ... | Developer Tools | 1m | 8 | Confirmed | ... |
| 3 | ... | SaaS | 3m | 7 | Watch | ... |

## GitHub Trending Watch

| Repository | Category | Window | Momentum | Why It Matters | Caveat |
|------------|----------|--------|----------|----------------|--------|
| owner/repo | AI/ML | 1w | +000 stars, 0 releases | ... | ... |
| owner/repo | Developer Tools | 1m | +000 stars, active issues | ... | ... |

## Model Reranking Watch

**Leaderboard**: LMArena / Artificial Analysis / LiveBench / SWE-bench / OpenRouter  
**Checked**: YYYY-MM-DD HH:MM TZ  

| Rank | Model | Provider | Metric | Leaderboard | Change | Core Reason |
|------|-------|----------|--------|-------------|--------|-------------|
| 1 | ... | ... | score / usage | ... | +0 | ... |
| 2 | ... | ... | score / usage | ... | -1 | ... |
| 3 | ... | ... | score / usage | ... | new | ... |
| 4 | ... | ... | score / usage | ... | 0 | ... |
| 5 | ... | ... | score / usage | ... | +2 | ... |
| 6 | ... | ... | score / usage | ... | ... | ... |
| 7 | ... | ... | score / usage | ... | ... | ... |
| 8 | ... | ... | score / usage | ... | ... | ... |
| 9 | ... | ... | score / usage | ... | ... | ... |
| 10 | ... | ... | score / usage | ... | ... | ... |

## Claude Agent Watch

| Signal | Area | Window | Status | Core Change | Action |
|--------|------|--------|--------|-------------|--------|
| ... | managed agents | 1w | Confirmed | ... | ... |
| ... | agent teams | 1m | Watch | ... | ... |

## Company Strategy Watch

| Company | Latest Move | Strategy Direction | Window | Status | Implication |
|---------|-------------|--------------------|--------|--------|-------------|
| OpenAI | ... | agents / models / apps / enterprise | 1m | Updated | ... |
| Anthropic | ... | agents / coding / enterprise / safety | 1m | Updated | ... |
| Alibaba Qwen | ... | models / open source / cloud API / enterprise | 1m | Updated | ... |
| Zhipu AI | ... | GLM models / agents / enterprise / domestic ecosystem | 1m | Updated | ... |
| Google Gemini/Gemma | ... | flagship models / open-weight models / ecosystem / distribution | 1m | Watch | ... |
| Microsoft/GitHub | ... | developer tools / enterprise | 1m | Watch | ... |

## Action Items

- [ ] ...
- [ ] ...
- [ ] ...

## Category Deep Dive

### AI/ML

#### [Signal Title]

- **Score**: 9/10
- **Status**: Confirmed / Watch / Rumor
- **Window**: 1w / 1m / 3m
- **What changed**: ...
- **Why it matters**: ...
- **Core change since last report**: new / updated / continuing / resolved
- **Evidence**: [Source name](URL), [Source name](URL)
- **Claude agent signal**: managed agents / agent teams / Claude Code / API primitive if relevant
- **Company strategy**: company, strategic direction, latest move if relevant
- **Model reranking**: leaderboard, rank, score/usage, change reason if relevant
- **GitHub signal**: repo link, stars/forks/releases if relevant
- **Impact window**: immediate / 1-3 months / 6-12 months
- **Watch next**: ...

### Developer Tools

#### [Signal Title]

- **Score**: 8/10
- **Status**: Confirmed
- **Window**: 1w / 1m / 3m
- **What changed**: ...
- **Why it matters**: ...
- **Core change since last report**: new / updated / continuing / resolved
- **Evidence**: [Source name](URL)
- **Claude agent signal**: managed agents / agent teams / Claude Code / API primitive if relevant
- **Company strategy**: company, strategic direction, latest move if relevant
- **Model reranking**: leaderboard, rank, score/usage, change reason if relevant
- **GitHub signal**: repo link, stars/forks/releases if relevant
- **Impact window**: ...
- **Watch next**: ...

## Weekly Stats

| Category | 1w Signals | 1m Signals | 3m Signals | Selected | Critical |
|----------|------------|------------|------------|----------|----------|
| AI/ML | 00 | 00 | 00 | 00 | 00 |
| Developer Tools | 00 | 00 | 00 | 00 | 00 |
| SaaS | 00 | 00 | 00 | 00 | 00 |

## Watchlist for Next Week

- [ ] ...
- [ ] ...
- [ ] ...

## Sources Reviewed

- [Source](URL) - short note
- [Source](URL) - short note

---

Generated by `tech-trend-radar`.

Writing Guidelines

• Lead with the conclusion, then provide evidence.
• Keep each signal detailed enough to be useful: one short paragraph plus 4-8 evidence bullets for major signals.
• Use tables for comparison, especially across 1w, 1m, and 3m windows.
• Include links for every important claim.
• For model reranking, include the leaderboard name, timestamp, metric definition, and current top 10.
• For OpenAI, Anthropic, Alibaba Qwen, and Zhipu AI strategy, include latest public moves, strategic direction, and implication in every monthly report.
• For Claude agent updates, prefer official Anthropic sources and extract the core change, affected users, and recommended action.
• Include GitHub repository links for open-source signals.
• If a topic was included in a previous report, still refresh it and write the newest delta.
• Avoid hype words unless the evidence supports them.
• Do not mix facts and speculation. Use Confirmed, Watch, or Rumor.
• When data is missing, say what is missing and how that affects confidence.

Output Location

Save reports to:

memory/research/tech-trend-radar-YYYY-WXX.md

If the user requests a different scope, include it in the filename:

memory/research/tech-trend-radar-ai-agents-YYYY-WXX.md
memory/research/tech-trend-radar-ai-agents-1m-YYYY-MM.md
memory/research/tech-trend-radar-ai-agents-3m-YYYY-QN.md

Cron Setup Example

Run every Monday at 09:00:

0 9 * * 1 openclaw run tech-trend-radar

Event Bus Integration

When a report is generated, publish:

events/tech-trend-radar-update-YYYY-MM-DD.json

Format:

{
  "type": "tech-trend-radar-report-generated",
  "timestamp": "2026-02-14T09:00:00Z",
  "week": "2026-W07",
  "window": "1w",
  "scope": ["AI/ML", "Developer Tools", "SaaS"],
  "highlights": [
    {
      "title": "...",
      "score": 9,
      "category": "AI/ML",
      "status": "Confirmed",
      "window": "1w"
    }
  ],
  "githubTrending": [
    {
      "repository": "owner/repo",
      "category": "Developer Tools",
      "window": "1w",
      "momentum": "+000 stars"
    }
  ],
  "modelReranking": {
    "leaderboard": "LMArena Text Arena Overall",
    "checkedAt": "2026-04-15T00:00:00Z",
    "top10": [
      {
        "rank": 1,
        "model": "...",
        "provider": "...",
        "metric": "..."
      }
    ],
    "coreChanges": ["..."]
  },
  "companyStrategy": [
    {
      "company": "OpenAI",
      "latestMove": "...",
      "strategyDirection": "...",
      "window": "1m",
      "status": "Updated",
      "implication": "..."
    },
    {
      "company": "Anthropic",
      "latestMove": "...",
      "strategyDirection": "...",
      "window": "1m",
      "status": "Updated",
      "implication": "..."
    },
    {
      "company": "Alibaba Qwen",
      "latestMove": "...",
      "strategyDirection": "...",
      "window": "1m",
      "status": "Updated",
      "implication": "..."
    },
    {
      "company": "Zhipu AI",
      "latestMove": "...",
      "strategyDirection": "...",
      "window": "1m",
      "status": "Updated",
      "implication": "..."
    }
  ],
  "claudeAgentSignals": [
    {
      "signal": "...",
      "area": "managed agents",
      "window": "1w",
      "status": "Confirmed",
      "coreChange": "...",
      "action": "..."
    }
  ],
  "signalsReviewed": 96,
  "signalsSelected": 12,
  "criticalSignals": 3
}

Customization

Edit keyword configuration at:

workspace/tech-trend-radar-keywords.json

Example:

{
  "AI/ML": ["OpenAI", "Claude", "Gemini", "Gemma", "Google Gemma", "Llama", "Qwen", "通义千问", "GLM", "智谱", "DeepSeek", "Kimi", "文心", "混元", "豆包", "AI agents", "SWE-bench", "model reranking", "Claude managed agents", "Claude agent teams"],
  "Company Strategy": ["OpenAI strategy", "Anthropic strategy", "Alibaba Qwen strategy", "阿里千问趋势", "通义千问", "Zhipu AI strategy", "智谱趋势", "GLM", "Google Gemini strategy", "Google Gemma strategy", "Gemma open model", "Microsoft AI strategy", "Meta AI strategy", "xAI strategy", "DeepSeek strategy", "Moonshot AI strategy", "Baidu ERNIE strategy", "Tencent Hunyuan strategy", "ByteDance Doubao strategy", "AI pricing", "enterprise AI", "consumer AI apps", "agent platform strategy"],
  "Claude Agents": ["Claude managed agents", "Claude agent teams", "Claude Code", "Anthropic agents", "MCP", "tool use", "computer use", "agent orchestration", "enterprise controls", "connectors"],
  "Model Reranking": ["LMArena", "Text Arena", "Artificial Analysis", "LiveBench", "SWE-bench", "OpenRouter rankings", "open-weight model leaderboard", "Gemma ranking", "Llama ranking", "Chinese model leaderboard", "Qwen ranking", "GLM ranking", "DeepSeek ranking", "Kimi ranking", "top 10 AI models"],
  "Developer Tools": ["Codex", "GitHub Copilot", "Cursor", "Claude Code", "Vercel", "LangGraph", "GitHub Trending"],
  "GitHub Trending": ["AI", "LLM", "agent", "RAG", "MCP", "devtools", "security", "TypeScript", "Python", "Rust", "Go"],
  "SaaS": ["Notion AI", "Linear", "Slack AI"],
  "Agent Platform Ecosystem": ["Anthropic", "Claude API", "Claude managed agents", "Claude agent teams", "Claude Code", "OpenAI", "Codex", "OpenAI Agents SDK", "Alibaba Qwen", "通义千问", "Zhipu GLM", "智谱", "Google Gemini", "Google Gemma", "Gemma", "Gemini agents", "GitHub Copilot", "Copilot coding agent", "MCP", "OpenClaw", "HarnessClaw"],
  "windows": ["1w", "1m", "3m"]
}

Keep keyword lists specific enough to reduce noise, but broad enough to catch new product names and adjacent market shifts.