Nanostack plugin
A 13-skill delivery workflow for coding agents: from idea to plan, review, security audit, tests and a shipped pull request
- 208 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The `init-project.sh` script pre-grants Claude Code permissions for files and commands in `.claude/settings.json`
- `/feature` mode approves the plan itself and drives the work to a pull request unless there are blocking findings
Install
- Works in
Manual install
npx create-nanostackDetects installed agents and sets up all skills. Needs `bash`, `git` and `jq`. Run `/nano-run` in the agent afterwards.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The `init-project.sh` script pre-grants Claude Code permissions for files and commands in `.claude/settings.json`
- `/feature` mode approves the plan itself and drives the work to a pull request unless there are blocking findings
This is third-party code. Review the repository files before installing.
What it does
Nanostack gives a coding agent a seven-phase sprint. `/think` refines a rough idea with questions and finds the smallest useful version, `/nano` plans the change with files, risks and checks. Then comes the build, followed by `/review` with two passes and scope-drift detection, `/security` with an OWASP A01-A10 audit and STRIDE threat model, `/qa` testing in a browser, API or CLI, and `/ship` with a pull request, CI check and sprint journal. Each phase writes its result to a local JSON file in `.nanostack/`, and the next phase reads those files rather than a chat recap. For autonomous runs there are `/think --autopilot` and `/feature <goal>`: once the plan is approved, the agent goes through all phases by itself. On Claude Code a guard layer built on `PreToolUse` hooks blocks dangerous commands and refuses `git commit` until fresh review, security and QA results exist. On Cursor, Codex, OpenCode and Gemini CLI the same rules run as instructions, without blocking.
Who it is for. People who hand whole features to a coding agent in an existing project and want a recorded plan, review and checks instead of one long answer.
Good fit when
- The idea is rough and has to be narrowed to a minimal version before a plan is built
- An agent builds a whole feature and you want separate review, security audit and tests before the commit
- Each phase's result should be saved as a file so you can later see what was decided and checked
- A team needs its own set of phases, for example a license check before release
- You would rather try the process on a sandbox like `starter-todo` without touching working code
Not a fit when
- A two-line fix: seven phases and result files cost more than the task
- You need hard blocking outside Claude Code: on other agents the guard works only as instructions
- The machine has no `bash`, `git` and `jq` and you do not want to set up WSL or Git Bash
- Autonomous mode is unwelcome: `/feature` approves the plan itself and continues to a pull request
Example request
/feature add a per-user rate limit check to the API with tests and updated docsLimitations
Hook-based blocking is verified only on Claude Code. On the other verified adapters (Cursor, Codex, OpenCode, Gemini CLI) the rules work as instructions, and `/nano-doctor` reports the real protection level of your install. It needs `bash`, `git` and `jq`, and on Windows runs through Git Bash or WSL.
The `init-project.sh` script creates `.claude/settings.json` with pre-granted permissions so Claude Code does not ask about every file and command. Telemetry is opt-in (`nanostack-config set telemetry anonymous`) and the project has no cloud of its own. The latest commit in the repository is from September 10, 2026.
How to disable. For Claude Code, delete the skill folders from `~/.claude/skills`: `think`, `nano`, `review`, `qa`, `security`, `ship`, `guard`, `conductor`, `compound` and `nanostack`. For Codex and OpenCode remove `~/.agents/skills/nanostack*`, for Cursor the `.cursor/rules/nanostack.md` file, and for Gemini CLI run `gemini extensions remove nanostack`.
FAQ
How do I install Nanostack in Claude Code?
Run `npx create-nanostack`: the installer detects agents, sets up all skills and runs the setup. Then type `/nano-run` in the agent to configure the project. The set updates with `/nano-update`.
Which agents are supported?
Verified adapters exist for Claude Code, Cursor, OpenAI Codex, OpenCode and Gemini CLI. Gemini CLI can also install it as an extension: `gemini extensions install https://github.com/garagon/nanostack --consent`.
What does the guard block?
On Claude Code it stops commands such as mass deletion, destroying git history, dropping databases and running remote code, and reading secrets. Read-only phases also block writes, and the commit waits for fresh review, security and test results.
How does `/feature` differ from a normal sprint?
In a normal sprint you approve the plan. `/feature <goal>` approves it itself and continues through the phases, stopping only for blocking review findings or critical vulnerabilities.
Can I add my own phase?
Yes, through custom skill stacks. The repository example is `compliance-release`, which combines several custom skills into a license, privacy and release gate. Details are in `EXTENDING.md`.
Discussion
No questions or reviews yet.
Related
PluginsEleven skills for Claude Code and Codex: test audit, git worktree cleanup, plan pages, screen recording transcripts and 1Password sign-in
A Claude Code plugin that hands a task to a local no_human and reviews a branch with a reviewer that never saw the coder's session
Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
Garry Tan's skill suite that turns Claude Code into a virtual team: product, architecture, design, review, QA and shipping
A 2.4 KB charter for CLAUDE.md and three on-demand skills: the agent stops guessing, verifies results and gates irreversible steps
A Claude Code plugin: 39 commands for PRs, review, commits and specs, nine expert roles and hooks, in eight language versions
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Nanostack (this card) | no | 208 | |
| gstack | no | 140,343 | Garry Tan's set that inspired this project: a virtual team without per-phase file artifacts and hook guarding |
| Superpowers | no | 313,096 | general development practices, without recorded phase results and commit blocking |
README in short
The documentation walks through a full sprint with examples, describes the Guided and Professional modes, the L0-L4 enforcement table per agent and the guard rules. The `examples/` folder holds sandbox projects built on one HTML file, Bash, Node and a static page, each with a ready prompt and success criteria.