The fire skill executes a written implementation plan task-by-task with strict, disciplined verification and no unauthorized improvisation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install fire
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install fire using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is fire?
The fire skill functions as the execution engine for an AI coding agent, acting on prepared technical recipes. Much like a kitchen line cook following an order ticket exactly as written, this skill takes a pre-defined implementation plan and processes it systematically, station by station. It ensures that no task leaves the pass unchecked and absolutely nothing is improvised on the live codebase.
Operating within the broader ecosystem of Openclaw Skills, fire enforces rigid preconditions such as working exclusively on dedicated feature branches or isolated worktrees, and establishing a green baseline before modifying code. It bridges the gap between high-level planning and concrete execution, offering both a solo execution mode and a brigade mode that orchestrates dedicated implementer subagents for complex tasks.
fire Use Cases
- When a written, discrete implementation plan or recipe output is ready for execution.
- When a developer or orchestrator triggers commands like "fire", "execute the plan", "build it", or provides a specific plan file.
- When executing multi-task developer blueprints that require strict task-by-task verification, testing, and atomic commits.
- When orchestrating specialized implementer subagents sequentially across a structured codebase ticket.
How fire Works
- Precondition Validation: The skill verifies a discrete plan exists, moves execution off the default branch into a feature branch or isolated worktree, and runs the test suite to establish a green baseline.
- Critical Ticket Review: Before touching code, the agent reviews the entire plan against the live codebase to spot stale assumptions, missing helpers, or structural contradictions early.
- Mode Selection: Selects Brigade Mode (dispatching fresh implementer subagents sequentially per task) or Solo Mode (executing inline task-by-task) based on plan size and subagent availability.
- Task Execution & Watch-It-Fail: Follows steps exactly, intentionally running watch-it-fail steps, reading test outputs, committing per task, and updating the plan file's checkboxes in real-time.
- Divergence Handling: Identifies structural gaps or broken verifications, prompting immediate halts and rollbacks to consult the plan author rather than attempting blind retries or design improvisation.
- The Finish: Runs the entire test suite one last time, then presents the user with explicit finish options: merge locally, push/open a PR, keep the branch, or discard the work.
fire Setup
To utilize this execution engine within your Openclaw Skills workspace, ensure your environment is initialized and your implementation plan is generated.
Ensure you have an active plan file (typically generated via a recipe skill), then trigger execution:
# Trigger the execution engine using explicit intents
openclaw fire
# Alternative natural language triggers managed by the framework
# "execute the plan"
# "build it"
Ensure that your testing framework is globally accessible within the environment, as the skill will run automated checks before and after execution.
fire Data Schema & Taxonomy
The fire skill organizes its runtime operations, task management, and output tracking according to a structured taxonomy:
| Data Component | Format / Structure | Purpose |
|---|---|---|
| Input Plan | Markdown File (.md) with Checklist Boxes |
Contains discrete tasks, specific validation instructions, and explicit commit boundaries. |
| Feature Branch | Git Branch Named fire/<plan>-<date> |
Isolates all execution side-effects away from the default repository branch. |
| Worktree Directory | Gitignoring Git-Worktree Folder | Provides temporary workspace isolation when native harness tools are unavailable. |
| Live Worklist | Updated Markdown Checkboxes (- [x]) |
Serves as the real-time state tracking mechanism for human operators and crash recovery. |
fire Advanced Features
- Brigade Mode Orchestration: Dispatches single-task, isolated implementer subagents sequentially, injecting full task context while isolating subagents from raw plan files.
- Atomic Commit-per-Task Enforcements: Automates individual task isolation by committing successful variations sequentially as outlined in the plan design.
- Rigid Divergence Guardrails: Intercepts missing modules or structural design mismatches, automatically reverting partial progress to preserve a clean working tree.
- Automated Worktree Cleanup: Safely handles environment teardown, removing temporary workspaces only during local merges or explicit work discards.
SKILL.md
Loading
Related Openclaw Skills
Memory Handoff
A structured AI agent skill designed to capture and extract durable, high-value technical knowledge at the end of a development session for long-term retention.
Mise
A design-first workflow skill that ensures all engineering tasks are fully designed, scoped, and user-approved before writing a single line of code.
plate
Plate is a prose hygiene and leak-scrubbing skill that ensures outward-facing text like PRs, READMEs, blogs, and commits is free of identity, system, and infrastructure data leaks before publishing.
Website Development Decision Assistant
An automated web-research and scoring agent that evaluates search trends, competitor landscapes, and monetization viability to generate HTML feasibility reports.
expedite
An AI-driven execution engine that systematically processes code audits and bug backlogs, resolving issues one-by-one with rigorous validation and isolated commits.
line-check
A read-only repository audit skill that evaluates codebase health across seven stations and generates a prioritized, action-oriented backlog sorted by leverage.
Featured*
