Reference
Glossary of canonical terms, CLI command reference, and answers to frequently asked questions.
Glossary
| Term | Definition |
|---|---|
| Work order | The user’s request describing what they want accomplished. Input to the system. |
| Plan | The comprehensive document the LLM produces by refining the work order. Covers requirements, architecture, risks, and success criteria. Output of the planning step. |
| Execution blueprint | The structured output of task generation. Contains all tasks organized into phases with dependency mappings. |
| Phase | A group of tasks within the execution blueprint. Tasks within a phase execute in parallel; phases execute in sequence. This is the unit of parallelism. |
| Task | An atomic unit of work within a phase. Has 1-2 skills, acceptance criteria, and dependencies. Executed by a sub-agent with clean context. |
| Sub-agent | A specialized AI agent that executes a single task with focused, clean context. Not the main conversation agent. |
| Skill | A harness-agnostic Agent Skill that implements one step of the workflow (e.g., st-create-plan, st-execute-blueprint). Skills load automatically when the user’s intent matches their description. |
| Harness | The AI assistant environment (Claude Code, Gemini CLI, GitHub Copilot, Codex, etc.) in which skills run. |
| Hook | A lifecycle callback (Markdown file) executed at a specific point in the workflow. Nine hooks are available: PRE_PLAN, POST_PLAN, PRE_PHASE, POST_PHASE, PRE_TASK_ASSIGNMENT, PRE_TASK_EXECUTION, POST_TASK_GENERATION_ALL, POST_EXECUTION, POST_ERROR_DETECTION. |
| Workspace | The .ai/strikethroo/ directory tree containing plans, archive, config, hooks, and templates. Created by init. |
| Archive | The archive/ subdirectory inside the workspace where completed plans are moved for historical reference. |
CLI Reference
Strikethroo has two distribution channels: the CLI (workspace bootstrapping) and the skills installer (workflow delivery). They are independently re-runnable; the only coupling point is the workspace schema version.
Workspace Initialization
npx strikethroo init --harnesses <harness>[,<harness>...] [options]
Creates the shared .ai/strikethroo/ directory (plans, archive, config, hooks, templates) and copies harness-specific artifacts (e.g., .claude/agents/ for Claude).
Required flag:
| Flag | Description |
|---|---|
--harnesses <list> | Comma-separated harness names. Controls which per-harness artifacts are copied. Accepted values: claude, gemini, opencode, codex, github, cursor. |
Optional flags:
| Flag | Description |
|---|---|
--destination-directory <path> | Target directory for the workspace. Defaults to the current working directory. |
--force | Overwrite all files without prompting, even if the user has customized them. Useful for CI/automation. |
File conflict detection: On re-run, init compares file hashes against .ai/strikethroo/.init-metadata.json. Unchanged files are updated silently; modified files trigger a unified-diff prompt. Use --force to bypass prompts.
Examples:
# Single harness
npx strikethroo init --harnesses claude
# Multiple harnesses
npx strikethroo init --harnesses claude,gemini,codex
# Target a different directory
npx strikethroo init --harnesses claude --destination-directory /path/to/project
# Force overwrite (automation)
npx strikethroo init --harnesses claude --force
Skill Installation
npx skills add e0ipso/strikethroo
Installs the Agent Skills that implement the three-step workflow. Skills are fetched from the repository’s tagged release via the .claude-plugin/plugin.json manifest.
Pin a specific version:
npx skills add e0ipso/strikethroo@<tag>
Update skills:
Re-run the same command to pull the latest version.
Skill Removal
npx skills remove e0ipso/strikethroo
Removes the installed Agent Skills. The .ai/strikethroo/ workspace, plans, and configuration are not affected.
Shipped Skills
| Skill | Purpose |
|---|---|
st-create-plan | Strategic plan creation with mandatory clarification gates. |
st-generate-tasks | Task decomposition with dependency mapping and skill assignments. |
st-execute-blueprint | Execution orchestration across all tasks in a plan. |
st-refine-plan | Plan refinement loop with interactive and autonomous clarification modes. |
st-execute-task | Single-task execution with dependency and status checks. |
st-full-workflow | End-to-end orchestration chaining all three steps. |
Frequently Asked Questions
Setup and Installation
Does Strikethroo require API keys or additional costs?
No. It works within your existing AI assistant subscriptions (Claude Pro/Max, Gemini, GitHub Copilot, Codex, Open Code). No API keys, no pay-per-token charges, no external service dependencies.
How long does setup take?
Under 30 seconds. Run npx strikethroo init --harnesses claude followed by npx skills add e0ipso/strikethroo, and the workspace is ready.
Does it work with existing projects?
Yes. The init command merges with existing project structures without breaking existing files. Hash-based conflict detection preserves your customizations on re-init.
Can I use multiple AI assistants on the same project?
Yes. Initialize with multiple harnesses (--harnesses claude,gemini,codex). All harnesses share the same plans, tasks, and configuration. Team members can use different harnesses while collaborating.
Workflow
What is the three-step workflow?
- Planning: The
st-create-planskill refines your work order into a comprehensive plan. - Task generation: The
st-generate-tasksskill decomposes the plan into an execution blueprint – atomic tasks organized into dependency-mapped phases. - Execution: The
st-execute-blueprintskill implements each task using sub-agents with focused context.
Each step produces files in .ai/strikethroo/plans/ for human review before the next step begins.
Do I have to use all three steps?
No. You can use only plan creation without task generation, generate tasks without executing, execute specific tasks manually, or skip steps that do not apply. The three-step workflow is recommended but not mandatory.
What if I want fully automated execution?
The st-full-workflow skill chains all three steps in a single invocation. It is best suited for well-defined features with clear scope. For complex features that need review between steps, use the manual step-by-step workflow.
How does Strikethroo relate to plan mode?
It augments plan mode rather than replacing it. The output of your assistant’s built-in plan mode is often a useful starting point – feed it into the st-create-plan skill for structured refinement.
Customization
Can I customize the workflow?
Yes. Nine lifecycle hooks, four templates, and project-context files are all editable Markdown. See the Customization Guide for examples.
What file formats does it use?
All configuration, plans, tasks, hooks, and templates are Markdown (.md) with YAML frontmatter where applicable. No proprietary formats.
Architecture
How does context isolation work?
Each step and each task runs with a focused context window. The planning step sees only the work order. Task generation sees only the plan. During execution, each sub-agent receives only the single task it is executing plus its declared dependencies – not the full plan or other tasks. This prevents the context window from growing unboundedly across a complex project.
What happens when a task fails during execution?
The POST_ERROR_DETECTION hook fires, enabling custom remediation logic. The task status is set to failed, and dependent tasks are not started. You can fix the issue and re-run execution.
How are tasks executed in parallel?
Tasks within the same phase have no mutual dependencies and execute concurrently via sub-agents. Phases themselves run in sequence, so a phase starts only after all tasks in the previous phase have completed.
Comparison with Other Tools
How does Strikethroo differ from API-based tools like Plandex or Claude Task Master?
API-based tools require separate service setup, API keys, and pay-per-token pricing. Strikethroo works within your existing subscription at no additional cost. It uses file-based configuration (editable Markdown) rather than API configuration, and most operations work offline.
When should I use plan mode instead of Strikethroo?
Use plan mode for simple tasks (fewer than 3 steps) with clear requirements where scope creep is not a concern and the AI can complete the work in one session. Use Strikethroo for complex multi-step projects, tight scope control, multi-session work, or when you need review gates between planning and execution.