hai-stack skill
hai-stack walks a software change through five steps, from judging the idea to recording results, with one skill per step.
- 396 ★ on the repository
- Last commit
Minimal access
Why this level
- The skills only reshape the agent's workflow; code edits are made with your permissions.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/hai-stackDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skills only reshape the agent's workflow; code edits are made with your permissions.
This is third-party code. Review the repository files before installing.
What it does
A set of 19 skills for Claude Code and Codex splits an iteration into five steps: judge the idea, design, build, verify, record. Each step has its own skill, including `hai-idea`, `hai-prd`, `hai-architecture`, `hai-tdd`, `hai-debug` and `code-review-and-quality`. Names need not be memorized: the user describes the snag in plain words and the matching skill takes over. Every skill returns something checkable, such as a conclusion with evidence, a reproduction, a defect list or an acceptance report. Three corrective skills, `geju`, `goudi` and `hai-razor`, can join at any step.
Who it is for. Developers who run tasks through Claude Code or Codex and want a verifiable result instead of a retelling.
Good fit when
- A feature idea looks doubtful and a verdict is needed before work starts.
- Production misbehaves for an unknown reason.
- A diff needs review before merging.
- Documentation drifts from the code and needs checking.
- Architecture makes rule changes painful and boundaries are unclear.
Not a fit when
- The task is small, the cause is known and a plain edit is enough.
- The repository would serve a commercial product, which the license forbids.
- The agent does not read `SKILL.md` files.
Example request
Production repeats requests when `RETRY_ENABLED=false`, while local runs are fine. Find the cause and change nothing.Limitations
The CC BY-NC 4.0 license forbids commercial use. Descriptions and examples are written in Chinese. `hai-debug` is marked experimental: on the benchmark task the agent solved it without the skill. Evaluation results in `evals/` show methods survived the skill merge but do not prove a quality gain. The author advises checking a skill's value on your own tasks first; the `evals/` directory shows how.
How to disable. Run `make unlink` in the cloned repository: only links pointing at it are removed. The repository folder can then be deleted.
FAQ
How do I install the skills?
Clone the repository and run `make link`: directories under `skills/` get links in `~/.agents/skills/` and `~/.claude/skills/`. For another agent, link or copy them into its skills directory.
How do I update the set?
Run `git pull && make link`. It is safe to repeat: existing links stay, new ones are added and links to retired skills are cleaned up.
What if only one skill is needed?
Copy or link the one directory from `skills/`, for example `skills/code-review-and-quality`. The rest are optional.
Can the skill structure be validated?
Yes, `make validate` checks structure, resources, entry examples and script syntax.
Which agents run the skills?
Most agents that read `SKILL.md` files, for example Claude Code and Codex.
Discussion
No questions or reviews yet.
Related
SkillsSix linked skills take a project from a raw idea to a validated build plan and a first draft in Claude Code and Codex.
lencx skills
lencx-skills
Two skills for Claude Code and Codex: keel works through architecture questions, coding-protocol keeps code edits inside the task
Skills for DeepSeek: short plan, verify and review loops, each with a hard stop condition and real command output as proof.
Eleven skills for Claude Code and Codex: test audit, git worktree cleanup, plan pages, screen recording transcripts and 1Password sign-in
A skill that finds AI slop in code: duplicate helpers, empty abstractions, weak tests. It cleans one pattern per commit.
iuhoay/skills collects agent skills: Vanilla Rails, Linear, stacked pull requests, plan challenging and tidy changes.
README in short
The README maps skills into a goal, skill, output table and explains the choice: a known small bug goes through TDD, an unclear failure is diagnosed first. A passed review does not authorize merging and acceptance does not authorize deployment.