Strikethroo plugin
Spec-driven development tuned for time to merge: you read and edit the plan, not the diff, and approve it before execution starts.
- 59 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skills execute tasks in the repository: they run commands, create branches and commit per phase.
Install
- Works in
Manual install
npx strikethroo init --harnesses claudeBootstraps the shared workspace; `npx skills add e0ipso/strikethroo` then installs the workflow skills.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skills execute tasks in the repository: they run commands, create branches and commit per phase.
This is third-party code. Review the repository files before installing.
What it does
Strikethroo splits work into two steps. The `/st-create-plan` command turns a work order into a plan a human reads and edits: correcting the plan costs one sentence, correcting a finished diff costs a review cycle. After approval `/st-execute-blueprint` decomposes the plan into atomic tasks with a dependency map and runs them, one commit per phase. Each step has clean context: the planner sees only the work order and a task executor only its own task. An optional review pass runs on a second harness found on the system so nobody checks their own work. The reviewer only detects problems and a human fixes them. Behavior is tuned with plain Markdown: hooks at eleven points of the process, plan templates and a project context file. `npx strikethroo serve` opens a local viewer for plans and tasks.
Who it is for. Developers and team leads who hand large changes to an agent and want to read a plan rather than a giant diff.
Good fit when
- A task spans several days and the plan must be agreed before the first line of code.
- A team wants its test commands and coding standards in every plan.
- The executor needs an independent check by another agent on the cumulative diff.
- You need plans, tasks and the dependency graph visible in a browser.
Not a fit when
- A one-line change where a plan is overkill.
- The agent must write code at once with no plan approval step.
- Node.js 22 or newer is unavailable.
Example request
Create a plan for work order 64: add SSO login. I will read the plan, edit it and only then start execution.Limitations
Node.js 22 or newer and an assistant with Agent Skills support are required. The workspace is created with `npx strikethroo init --harnesses claude` and lives in `.ai/strikethroo`. Skill scripts run commands in the repository, commit per phase and create a feature branch. The review pass needs a second agent on the system, otherwise it does not run. The viewer listens on `127.0.0.1:4317`, and exposing it to another machine takes the `--host` flag.
How to disable. Remove the skills through the agent skill manager and the `.ai/strikethroo` directory from the project. If you added the plugin, remove it with the Claude Code plugin uninstall command.
FAQ
How do I install Strikethroo?
Run `npx strikethroo init --harnesses claude`, then `npx skills add e0ipso/strikethroo`. `npx strikethroo@latest update` updates it, and a fresh agent session is needed afterwards.
Which commands does the process have?
`/st-create-plan` creates a plan and `/st-execute-blueprint` builds tasks and runs them. A separate `/st-generate-tasks` is only needed to hand-tune the task blueprint before execution.
How do I tailor it to my project?
Add hooks, templates and a project context file as Markdown. Hooks fire at eleven points: before planning, after phases and on errors.
What is a profile?
A packaged configuration with hooks, templates and context: `init --profile` starts from it and `export profile` packages yours for sharing.
How do I view plans in a browser?
Run `npx strikethroo serve` in the project. The page opens on `127.0.0.1:4317` and shows current and archived plans with tasks.
Discussion
No questions or reviews yet.
Related
PluginsSkills for spec-driven development: brainstorming, planning, debate-style review and idea grilling before the first commit.
A plugin with three skills for large development work: project analysis, phased tasks tracked as GitHub Issues, parallel agents and code review
Orchestrator skill: from an idea the agent writes requirements and a spec, splits work into tasks, builds the project, checks it
planr keeps agent-driven development tasks in Git Markdown files and shows what is ready, blocked and waiting on you
A planning extension for the Pi coding agent: reviewed Markdown plans first, and code changes only after your approval
HCF separates planning with a human from autonomous execution: parallel workers implement the plan using TDD.
README in short
The authors assume an agent is an unreliable judge of its own work and that the scarce resource is human attention rather than tokens. So review happens on the plan, guarantees are compiled into code and tuning needs no plugins.