alamo-skills

Alamo Skills

English | 简体中文

What Is In Here

All four are Apache-2.0, pure Python 3.8+ standard library plus pyyaml, and carry no backend.

The Packages

coursecoach

The odd one out — it is not about code, it is a learning coach. Give it a topic and a daily time budget and it produces a course that runs: a dependency-ordered syllabus, a compiled interactive board you double-click, written lessons, gradable homework, mastery-gated unlocking, and spaced review.

It is subject-agnostic — programming, a foreign language, an instrument, accounting, an exam all work. What separates it from "ask a model for a study plan" is what happens after the plan: scores, unlocks and review dates live in the files, so a course survives across sessions, can be picked up by a different model, and can be zipped and handed to someone who just double-clicks the board.

Six things it does. All triggered in plain language, and every one has a matching button on the board that assembles the sentence for you:

• Start a course — say "I want to learn X, one hour a day." It asks about your goal, your current level and your real weekly capacity, offers three pace options (sprint / standard / long), then researches the field, builds the prerequisite graph over a dozen or so nodes, and compiles the board. Produces syllabus.md (the only source of truth), 课程主页.html, and resources/资料库.md.
• Write lessons — click 📖 补这一章讲解 (fill in this chapter) under any stage, or open a knowledge-point tag and pick 生成讲解. Works per chapter or per single point. Produces lessons/stage3.md or lessons/stage3--梯度下降.md; after a recompile the board grows a 📄 button that opens it typeset, with no Obsidian required.
• Set homework — say "give me homework on stage 3", or click 📝 给这一章出作业. Writes the questions plus 3–6 rubric lines and leaves the answer section empty. Produces homework/hw-*.md, always in three parts: prompt / answer / feedback.
• Grade it — answer the questions, then say "grade hw-stage3-xxx." It judges each rubric line present / partial / absent with a reason (fixed rubric, and length earns nothing extra), converts that to mastery, writes it back, unlocks dependent nodes, and queues the work for review.
• Fix weak spots — say "I keep failing at X in stage 3." It reads which knowledge points repeatedly came back partial or absent, decides whether to add resources, split a node or introduce a new one, and writes targeted homework if that is what the gap needs. Changes go through syllabus.py, never by hand-editing.
• Review and export — "what's due today" returns items on the 1 / 3 / 7 / 16 / 35 / 75-day curve (postpone without corrupting the count); "export to Word" gives you the whole course as one document.

Not installing anything yet? The live demo is a real course compiled from the example snapshot, with five lessons inlined: expand a stage, open a lesson, try the search. Nothing to set up.

Install it (three steps, about a minute):

git clone https://github.com/Alamoooooooo/alamo-skills.git
cp -r alamo-skills/packages/coursecoach/coursecoach ~/.workbuddy/skills/coursecoach
pip install pyyaml

• The installable unit is the directory holding SKILL.md, i.e. coursecoach/coursecoach/. Once copied, the folder name must be coursecoach — matching the skill's own name — or the slash command will not resolve.
• On Windows the destination is C:\Users\<you>\.workbuddy\skills\coursecoach\. Dropping it in a project's .workbuddy/skills/ scopes it to that project instead.
• Codex reads $CODEX_HOME/skills; hosts without skill support can take SKILL.md plus references/ straight into context and run the scripts by hand.
• Requires Python 3.8+, and that interpreter must be able to import yaml. Word export additionally needs pip install python-docx.
• Reopen the session before it is recognized. Then say "I want to learn X" or /coursecoach directly.

Do not start with a real subject — the install guide includes a 15-minute trial course that shows the whole loop before you commit to anything.

Read: package README · 中文说明 · install and 15-minute trial · portable usage

engineering_subagent_workflow_kernel

Five development skills that delegate through the host's own subagent mechanism instead of an external runner binary: code-explorer, code-architect, code-reviewer, code-simplifier, pr-test-analyzer.

Each skill is self-contained — its role reference ships with it. It does not install helper binaries, spawn nested agent processes, choose a model, or weaken the host's approval and sandbox policy. Hosts without native subagents fall back to in-session execution and report fallback_in_session so the difference is visible.

Read: package README · portable usage

cd packages/engineering_subagent_workflow_kernel
python3 scripts/sync_to_codex_home.py --dry-run
python3 scripts/sync_to_codex_home.py

github_sync_workflow_kernel

For the step everyone gets wrong once: deciding which directory is the real repository, whether the remote and auth are actually working, which changes belong in this commit and which are line-ending noise, and whether the push landed.

github-sync-agent runs repository discovery, auth and remote inspection, commit-scope review, commit, push, and post-push verification as one workflow. The wiki documents the failure modes it was built against.

Read: package README · wiki · portable usage

cd packages/github_sync_workflow_kernel
python3 scripts/sync_to_codex_home.py

memory_hotspot_repair_kernel

For Python services whose RSS keeps climbing, batch jobs that spike during joins or model scoring, and pipelines where "chunking" still loads the whole dataset.

Four coordinated skills: memory-check inspects and ranks hotspots, memory-fix remediates the highest-value one, memory-review reviews the change independently, and memory-optimizer-agent orchestrates all four into one bounded loop with explicit stop conditions. The other three also run standalone.

Read: package README · long-form wiki · portable usage

cd packages/memory_hotspot_repair_kernel
python3 scripts/sync_to_codex_home.py

Pick Your Starting Point

Install

Codex

git clone https://github.com/Alamoooooooo/alamo-skills.git
cd alamo-skills/packages/memory_hotspot_repair_kernel
python3 scripts/sync_to_codex_home.py

sync_to_codex_home.py copies into $CODEX_HOME/skills, or ~/.codex/skills when CODEX_HOME is unset. Add --dry-run to preview. The three kernel packages each have their own copy of this script; install only the ones you want.

WorkBuddy, CodeBuddy, Claude Code, or any host that reads SKILL.md

git clone https://github.com/Alamoooooooo/alamo-skills.git
cp -r alamo-skills/packages/coursecoach/coursecoach ~/.workbuddy/skills/coursecoach

The installable unit is always the directory that contains SKILL.md — coursecoach/coursecoach/, memory-optimizer-agent/, github-sync-agent/, and so on. pip install pyyaml is required by coursecoach; export.py --format docx additionally needs python-docx.

For hosts that understand neither skill installation nor $name tokens, read the relevant PORTABLE_USAGE.md: it explains how to feed the manual and run the scripts by hand.

Repository Layout

alamo-skills/
├── README.md                          English (this file)
├── README.zh-CN.md                    简体中文
├── CONTRIBUTING.md                    how to change this repo: checks, commits, review
├── CONTRIBUTING.zh-CN.md              简体中文
├── AGENTS.md                          conventions for agents working on this repo itself
├── llms.txt                           machine-readable summary for AI crawlers
├── LICENSE                            Apache-2.0
├── scripts/check.py                   repo self-check: pairing, dead links, metadata, syntax
├── .github/pull_request_template.md   the checklist a PR opens with
├── packages/                          ← the products
│   ├── README.md                      package conventions: three layers, required files, references vs wiki
│   ├── coursecoach/
│   │   ├── coursecoach/               the skill itself (SKILL.md, references, scripts, examples)
│   │   ├── docs/                      board.png, lesson-reader.png
│   │   └── README.md  README.zh-CN.md  PORTABLE_USAGE.md  INSTALL.md  INSTALL.zh-CN.md  metadata.json
│   ├── engineering_subagent_workflow_kernel/
│   │   ├── code-explorer/  code-architect/  code-reviewer/
│   │   ├── code-simplifier/  pr-test-analyzer/
│   │   ├── scripts/
│   │   └── README.md  INSTALL.md  INSTALL.zh-CN.md  PORTABLE_USAGE.md  metadata.json
│   ├── github_sync_workflow_kernel/
│   │   ├── github-sync-agent/
│   │   ├── scripts/  wiki/
│   │   └── README.md  INSTALL.md  INSTALL.zh-CN.md  PORTABLE_USAGE.md  metadata.json
│   └── memory_hotspot_repair_kernel/
│       ├── memory-optimizer-agent/    orchestrator (primary entry point)
│       ├── memory-check/  memory-fix/  memory-review/
│       ├── references/  scripts/  wiki/
│       └── README.md  INSTALL.md  INSTALL.zh-CN.md  PORTABLE_USAGE.md  metadata.json
├── examples/                          ← optional, never required to install
│   ├── coursecoach/
│   │   └── quick-trial/stages.json    the smallest input that produces a 3-node course
│   └── memory_hotspot_repair_kernel/
│       └── partitioned_batch_pipeline/    one repo-specific adaptation: overlays + run configs
└── todo/                              ← unimplemented plans, nothing depends on them
    └── codex-plugin-packaging.md

packages/ is the product, examples/ is optional, todo/ is ideas nobody has promised to build. The conventions that govern a package — the three layers, which files every package carries, what separates references/ from wiki/, and which directory you are actually supposed to copy — live in packages/README.md.

Other Files

License

Apache-2.0. See LICENSE.