๐Ÿฟ๏ธ Squirrel

April 29, 2026 ยท View on GitHub

๐Ÿฟ๏ธ Squirrel

License Platforms Agent Skills Spec

All Contributors

Full-cycle AI coding skill that works everywhere โ€” from blank canvas to production.

Plans, builds, tests, lints, fixes bugs, and writes production-grade docs. Runs on OpenCode, Codex, Claude Code, Cursor, Windsurf, Aider, Cline, Copilot, and Antigravity.

Install ยท How it works ยท Report Bug


Features

  • Auto-detects project state โ€” Greenfield, in-progress, or mature. Squirrel figures out where your project is and jumps in at exactly the right point instead of forcing a one-size-fits-all workflow.
  • Respects existing code โ€” Matches your naming conventions, test framework, import style, and architecture. Extends what's there instead of overwriting it. Reads 2โ€“3 similar files before writing a new one.
  • 8-phase engineering pipeline โ€” Discover โ†’ Plan โ†’ Build โ†’ Test โ†’ Bug Hunt โ†’ Polish โ†’ Document โ†’ Ship. The same disciplined process a senior engineer follows, every time.
  • Platform-agnostic โ€” One skill file, 9 AI coding agents. Drop it in as AGENTS.md, CLAUDE.md, .cursor/rules/squirrel.mdc, or any of the other supported formats and it just works.
  • Built-in failure recovery โ€” 3-Strike Rule: fix, retry differently, then stop and ask. Never leaves code in a broken state. Includes sub-agent failure recovery for parallel execution.
  • CI/CD templates included โ€” Ready-to-use GitHub Actions workflows for Node.js, Python, Go, Rust, and more. Use them as a starting point, not a drop-in guarantee.
  • Stack-agnostic โ€” Works with TypeScript, Python, Go, Rust, Ruby, Java, C#, Elixir, and any language that has a linter, formatter, and test runner. Detects your stack and adapts automatically.

Quick Start

Install with skills.sh (works with 50+ AI agents):

npx skills add flyingsquirrel0419/squirrel-skill

Or install in one command โ€” auto-detects your AI agent and platform:

curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/install.sh | bash

Or install for a specific platform:

curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/install.sh | bash -s -- --platform cursor

Or clone and run locally:

git clone https://github.com/flyingsquirrel0419/squirrel-skill.git
cd squirrel-skill && bash install.sh

Then just tell your AI agent what you want to build:

> build me a REST API for a todo app with TypeScript and Express
> create a CLI tool in Rust that parses CSV files
> fix this bug in src/auth/login.py
> squirrel this project โ€” add tests, fix lint errors, write README

Squirrel auto-detects your project state and starts at the right phase. No config needed.


Installation

skills.sh (works with 50+ agents)

npx skills add flyingsquirrel0419/squirrel-skill

Universal installer โ€” auto-detects your agent and installs Squirrel with references. No setup needed.

curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/install.sh | bash

Auto-detects which AI agents are configured in your project and installs Squirrel for all of them.

Install for a specific platform

# Any platform
bash install.sh --platform <platform>

# Examples
bash install.sh --platform opencode
bash install.sh --platform cursor
bash install.sh --platform claude-code
bash install.sh --platform aider

# Custom output path
bash install.sh --platform codex --path ./my-instructions.md

Supported platforms: opencode, codex, claude-code, cursor, windsurf, aider, cline, copilot, antigravity

Manual install per platform

OpenCode
mkdir -p ~/.config/opencode/skills/squirrel
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md \
  -o ~/.config/opencode/skills/squirrel/SKILL.md
OpenAI Codex
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md -o AGENTS.md
Claude Code
# Option A: Direct
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md -o CLAUDE.md

# Option B: Import from existing CLAUDE.md
echo -e "\n@AGENTS.md" >> CLAUDE.md
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md -o AGENTS.md
Cursor
mkdir -p .cursor/rules
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md -o .cursor/rules/squirrel.mdc
# Then add frontmatter to squirrel.mdc:
# ---
# description: Squirrel full-cycle development skill
# alwaysApply: true
# ---
Windsurf
mkdir -p .windsurf/rules
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md -o .windsurf/rules/squirrel.md
# Then add frontmatter to squirrel.md:
# ---
# trigger: always_on
# description: Squirrel full-cycle development skill
# ---
Aider
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md -o squirrel-skill.md
aider --read squirrel-skill.md
# Or add to .aider.conf.yml:  read: squirrel-skill.md
Cline
mkdir -p .clinerules
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md -o .clinerules/squirrel.md
GitHub Copilot
mkdir -p .github
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md \
  -o .github/copilot-instructions.md
Antigravity
mkdir -p ~/.gemini/antigravity/skills/squirrel
curl -fsSL https://raw.githubusercontent.com/flyingsquirrel0419/squirrel-skill/main/skills/squirrel/SKILL.md \
  -o ~/.gemini/antigravity/skills/squirrel/SKILL.md

How It Works

Step 0: Detect Mode

Squirrel checks your project directory and classifies it:

SignalModeEntry Point
Empty directory, no source files๐Ÿ†• GreenfieldAll 8 phases from scratch
Source files, no tests/docs๐Ÿ”ง In-ProgressAudit first, then improve
Source + tests + CI + README๐Ÿ—๏ธ MatureTargeted improvements
"fix this bug / add this feature"๐ŸŽฏ TargetedAbbreviated audit, scoped work

The 8-Phase Pipeline

[1] Discover  โ†’ Understand the project (audit existing code or gather requirements)
[2] Plan      โ†’ Concrete task list with dependencies, risks, and done-criteria
[3] Build     โ†’ Write or modify code (parallel sub-agents when platform supports it)
[4] Test      โ†’ Run existing tests, write new ones, 70%+ coverage target
[5] Bug Hunt  โ†’ Static analysis + manual review, fix all critical/warning bugs
[6] Polish    โ†’ Lint, format, type check, remove dead code
[7] Document  โ†’ README + inline docs (update existing, don't overwrite)
[8] Ship      โ†’ Final checklist: tests green, no secrets, CI configured, clean checkout

Failure Recovery (3-Strike Rule)

  1. Strike 1: Fix the specific error. Run tests. Move on.
  2. Strike 2: Re-read the code. Try a different approach.
  3. Strike 3: STOP. Revert. Document what failed. Ask the user.

Code is never left in a broken state. Failing tests are never deleted to "pass".

Parallel Execution

On platforms that support sub-agents (OpenCode, Codex, Cursor), Squirrel decomposes work into independent units and spawns them in parallel using a structured delegation prompt:

TASK: [atomic goal]
CONTEXT: [relevant plan sections, shared types]
SCOPE: [files to modify, files NOT to touch]
DONE WHEN: [verifiable success criteria]
STYLE: [existing patterns to follow]
CONSTRAINTS: [what not to do]

Reference Files

Squirrel ships with supplementary references loaded on demand:

FilePurpose
references/plan_template.mdProject plan template with risk matrix, task breakdown, progress log
references/readme_template.mdProduction-grade README template with badges, quick start, config table
references/stack_hints.mdCommon pitfalls and best practices for TS, Python, Go, Rust, React, REST, DB
references/ci_templates.mdGitHub Actions CI/CD for Node.js, Python, Go, Rust + npm/PyPI release workflows

Development

Project structure

squirrel/
โ”œโ”€โ”€ .all-contributorsrc              # All Contributors config for README credits
โ”œโ”€โ”€ .github/workflows/ci.yml         # Smoke-test workflow for installer regressions
โ”œโ”€โ”€ install.sh                       # One-liner installer with auto-detection
โ”œโ”€โ”€ skills/
โ”‚   โ””โ”€โ”€ squirrel/
โ”‚       โ”œโ”€โ”€ SKILL.md                 # Main skill definition
โ”‚       โ””โ”€โ”€ references/
โ”‚           โ”œโ”€โ”€ ci_templates.md      # CI/CD pipeline templates
โ”‚           โ”œโ”€โ”€ plan_template.md     # Project plan template
โ”‚           โ”œโ”€โ”€ readme_template.md   # README template
โ”‚           โ””โ”€โ”€ stack_hints.md       # Language-specific best practices
โ””โ”€โ”€ tests/
    โ””โ”€โ”€ installer_smoke_test.sh      # Verifies installer output and doc slug consistency

Contributing

  1. Fork the repo
  2. Create a branch: git checkout -b feature/your-feature
  3. Edit skills/squirrel/SKILL.md or add references in skills/squirrel/references/
  4. Test your changes on at least one supported platform
  5. Push and open a PR

Design principles

  • One file does everything. SKILL.md is the single source of truth. Reference files are supplementary, not required.
  • No platform assumptions. Every instruction must make sense on all 8 platforms. Platform-specific details go in the Platform Compatibility section only.
  • Respect > Rewrite. The skill teaches the agent to respect existing code. The skill itself follows the same principle โ€” extend, don't replace.
  • Plain Markdown is portable. No proprietary syntax, no code-only constructs. Every AI agent reads Markdown.

Contributors

This project follows the All Contributors specification. Contributions of any kind are welcome.

flyingsquirrel0419
flyingsquirrel0419

๐Ÿ’ป ๐Ÿ“–

License

Apache 2.0 ยฉ flying_squirrel__