Altinity

codex-summarize-pipeline

2
1
# Install this skill:
npx skills add Altinity/Skills

Or install specific skill: npx add-skill https://github.com/Altinity/Skills/tree/main/experimental/codex-summarize-pipeline

# Description

Summarize huge articles (URL or local file) via a Codex CLI-driven chunk→reduce pipeline, keeping only the final short summary in context and saving it to summaries/*.md.

# SKILL.md


name: codex-summarize-pipeline
description: Summarize huge articles (URL or local file) via a Codex CLI-driven chunk→reduce pipeline, keeping only the final short summary in context and saving it to summaries/*.md.


Codex Summarize Pipeline

Use this skill to summarize long articles without loading the full text into the chat context.

Guarantees / constraints

  • Use only local tools and the codex CLI (no direct API calls in your code).
  • Do not paste the full article into the conversation.
  • Keep only the final short summary (and the output file path) in the assistant response.
  • Always save the summary to a markdown file under summaries/.

See README.md in this skill for detailed usage and recommended defaults for HTML-extracted blogs/docs.

Default flow (URL)

  1. Pick a slug for filenames (short, filesystem-safe).
  2. Fetch + extract text:
  3. python3 <skill>/tools/fetch_article_text.py --url "<URL>" --output .codex-pipeline/inputs/<slug>.txt
  4. Summarize (default map-reduce):
  5. python3 <skill>/tools/codex_summarize_pipeline.py --input .codex-pipeline/inputs/<slug>.txt --final-max-words 120 > summaries/<slug>.summary.md
  6. Respond with:
  7. The path summaries/<slug>.summary.md
  8. The final summary content only (short)

Default flow (local file)

  • python3 <skill>/tools/codex_summarize_pipeline.py --input <path> --final-max-words 120 > summaries/<slug>.summary.md

Modes

  • map-reduce: best general-purpose coverage.
  • extract-synthesize: best when names/numbers/timeline matter.
  • question-led: best when user has a specific goal (use --goal).

Troubleshooting

  • If extraction looks wrong, save the HTML and rerun:
  • python3 <skill>/tools/fetch_article_text.py --input-html page.html --output .codex-pipeline/inputs/<slug>.txt
  • If codex is configured to run remote models and you want local-only, run with --oss + --local-provider (requires local LM server).

# README.md

Skills

This repository contains skills used for ClickHouse DB performance and schema analysis and helper workflows.

Core Skills

  • altinity-expert-clickhouse/: Modular ClickHouse diagnostic skill set. Each module is a standalone skill under altinity-expert-clickhouse/skills/ (e.g., memory, merges, replication).
  • automations/: Batch and focus audit scripts that run full ClickHouse analysis and emit a single report.
  • releases/: Built zip packages for distribution (one per skill).

Install / Use Skills

1) Codex CLI

variants:
- use $skill-installer skill inside codex
- clone repo and copy needed skills into ~/.codex/skill directory.
- clone repo and ln (symlink) repo's skill directory into ~/.codex

2) Claude CLI (Claude Code)

variants:
- claude skills add URL
- clone repo and copy needed skills into ~/.claude/skill directory.
- clone repo and ln (symlink) repo's skill directory into ~/.claude

3) Gemini CLI

Install an agent skill from a git repository URL or a local path.

gemini skills install <source> [--scope] [--path] 

or manually:

git clone https://github.com/Altinity/Skills.git
mkdir -p ~/.gemini/skills
cp /absolute/path/to/Skills/altinity-expert-clickhouse/skills/altinity-expert-clickhouse-memory ~/.gemini/skills/
# or
ln -s /absolute/path/to/Skills/altinity-expert-clickhouse/skills ~/.gemini/skills/

4) Claude.ai (web)

Download the zip files from releases/ and upload them in Settings (or Admin Settings for org‑wide use) to Capabilities section.

Experimental Skills

  • experimental/codex-summarize-pipeline/: Chunk→reduce pipeline for summarizing large articles/files into summaries/*.md.
  • experimental/github-triage/: Search and summarize relevant GitHub issues/PRs using gh.
  • experimental/sub-agent/: Prototype sub-agent workflow (multi-agent attempt by exec of next codex/claude inside skill processing).

Conventions

  • Each skill lives in its own directory and includes a SKILL.md.
  • Supporting content is stored next to SKILL.md (e.g., modules/, scripts, prompts).

# Supported AI Coding Agents

This skill is compatible with the SKILL.md standard and works with all major AI coding agents:

Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.