Trender Skill

Trender is an agent-native skill for mapping how a topic changes over time. It combines bucketed web evidence from the host coding agent with a bundled last30days community-research layer, then clusters themes, scores momentum, surfaces emerging entities and vocabulary drift, and renders a self-contained HTML trend map.

Current version: 0.7.0.

What it does

• Compares flexible time windows, including --compare=7,30, --compare=30,180, --days, or explicit --from / --to ranges.
• Uses host-agent web research via --agent-web-file for higher-quality evidence.
• Bundles a compatible last30days engine in skills/trender/vendor/last30days.
• Routes community subqueries by intent instead of sending every query to every source.
• Produces inflection moments, accelerating/fading/stable themes, then-to-now quote pairs, emerging entities, vocabulary drift, forward signals, an agent-authored BLUF/forward outlook when provided, Markdown synthesis, JSON data, and an HTML report.
• Opens the HTML report automatically by default.

Trender itself does not call OpenAI, Brave, or similar web APIs directly. For deep web research, the host agent should use its own web/deep-research tools, save evidence to JSON, and pass that file to Trender. Optional credentials can improve the bundled last30days sources.

Requirements

• Python 3.12+ for the bundled last30days community layer.
• Shell access from the host agent.
• Optional network access and credentials for richer source coverage.

Useful environment variables:

Variable	Purpose
TRENDER_OUTPUT_DIR	Override the default report directory, ~/Documents/Trender.
TRENDER_AGENT_WEB_FILE	Default --agent-web-file path.
LAST30DAYS_SKILL_DIR	Override the bundled last30days engine.
TRENDER_LAST30DAYS_PYTHON	Python executable to use for last30days when multiple versions are installed.

Agent-authored narrative

The host coding agent can provide a bottom-line-up-front (BLUF) summary and forward outlook with --narrative-file. This lets the final report lead with the agent's synthesized takeaways while Trender keeps the evidence-backed trend map, computed signals, and traceable source lists.

If no narrative file is provided, Trender renders a clearly labeled auto-generated fallback based on the strongest computed signals.

Repository layout

plugin.json                    GitHub Copilot CLI plugin manifest
.claude-plugin/plugin.json      Claude Code plugin manifest
.codex-plugin/plugin.json       Codex plugin manifest
LICENSE                        MIT license included in built archives
skills/trender/SKILL.md         Agent skill instructions and metadata
skills/trender/README.md        README packaged inside the skill archive
skills/trender/scripts/         CLI, installer, and build scripts
skills/trender/vendor/          Bundled last30days engine
dist/trender.skill              Built direct Agent Skill archive
dist/trender-plugin.zip         Built plugin archive

Install as an agent skill

Install as a personal GitHub Copilot CLI skill on Windows:

powershell -NoProfile -ExecutionPolicy Bypass `
  -File .\skills\trender\scripts\install-skill.ps1 `
  -Agent copilot `
  -Force

Install for Claude Code or Codex on Windows:

powershell -NoProfile -ExecutionPolicy Bypass `
  -File .\skills\trender\scripts\install-skill.ps1 `
  -Agent claude `
  -Force

powershell -NoProfile -ExecutionPolicy Bypass `
  -File .\skills\trender\scripts\install-skill.ps1 `
  -Agent codex `
  -Force

Default install locations:

GitHub Copilot CLI: %USERPROFILE%\.copilot\skills\trender
Claude Code:        %USERPROFILE%\.claude\skills\trender
Codex / agents:     %USERPROFILE%\.agents\skills\trender

On macOS/Linux:

bash ./skills/trender/scripts/install-skill.sh --agent copilot
bash ./skills/trender/scripts/install-skill.sh --agent claude
bash ./skills/trender/scripts/install-skill.sh --agent codex

After direct skill installation, reload or restart the host if needed:

Copilot CLI: /skills reload
Claude Code: /reload-plugins or restart
Codex: restart if the skill is not detected automatically

Install as a plugin

For plugin distribution, install the repository root:

# GitHub Copilot CLI
copilot plugin install .

# Claude Code local plugin test
claude --plugin-dir .

Codex can install from a marketplace entry that points at this repository or local plugin folder. The Codex manifest is in .codex-plugin/plugin.json.

Configure sources

Trender works without separate setup in --mock mode and can use free bundled last30days sources such as Reddit, Hacker News, Polymarket, and GitHub when available.

Run diagnostics:

python .\skills\trender\scripts\trender.py --diagnose

Run setup once to let the bundled last30days layer discover browser cookies and write its local environment file:

python .\skills\trender\scripts\trender.py setup

Add host-agent web evidence

For best results, have the host coding agent gather bucketed evidence first. Use separate web searches for each bucket: research, implementations, adoption, criticism, and forecasts. Each item should include a concrete published_at date because undated evidence cannot contribute to trend analysis.

{
  "buckets": {
    "research": [
      {
        "title": "...",
        "url": "https://...",
        "published_at": "YYYY-MM-DD",
        "summary": "...",
        "source": "arxiv",
        "relevance_score": 0.9
      }
    ],
    "implementations": [],
    "adoption": [],
    "criticism": [],
    "forecasts": []
  }
}

The legacy {"items":[...]} shape is still accepted, with buckets inferred from content. See skills/trender/SKILL.md Step 0 for the full evidence contract.

Then run:

python .\skills\trender\scripts\trender.py "MCP servers" --agent-web-file .\agent-web.json

Run

HTML is the default output and opens automatically:

python .\skills\trender\scripts\trender.py "MCP servers" --days=90

Use --no-open for scripts, CI, or unattended runs:

python .\skills\trender\scripts\trender.py "MCP servers" --days=90 --no-open

Other examples:

# Default when no window is specified: compare last 30 days vs prior 5 months.
python .\skills\trender\scripts\trender.py "agentic AI" --agent-web-file .\agent-web.json

# Include an agent-authored BLUF and forward outlook.
python .\skills\trender\scripts\trender.py "agentic AI" --agent-web-file .\agent-web.json --narrative-file .\narrative.json

# Compare two lookback windows.
python .\skills\trender\scripts\trender.py "agentic AI" --compare=7,30 --agent-web-file .\agent-web.json
python .\skills\trender\scripts\trender.py "AI video tools" --compare=30,180 --emit=all

# Analyze an explicit date range.
python .\skills\trender\scripts\trender.py "AI coding agents" --from=2026-01-01 --to=2026-06-01

# Network-free smoke test.
python .\skills\trender\scripts\trender.py "MCP servers" --mock --emit=all --no-open

By default, reports are written to ~/Documents/Trender:

<topic>-trend-map.html
<topic>-trend-map.json

Set TRENDER_OUTPUT_DIR or pass --save-dir to change the destination.

Build skill archives

powershell -NoProfile -ExecutionPolicy Bypass -File .\skills\trender\scripts\build-skill.ps1

On macOS/Linux:

bash ./skills/trender/scripts/build-skill.sh

Archives are written to:

dist\trender.skill       Direct Agent Skill archive
dist\trender-plugin.zip  Plugin archive with Copilot, Claude, and Codex manifests

Public release checklist

Before publishing, confirm the root LICENSE matches the MIT license declared in the manifests, rebuild dist/ after documentation or code changes, and run a mock smoke test:

python .\skills\trender\scripts\trender.py "MCP servers" --mock --emit=all --no-open