Develop Full-Cycle Software Projects

Squirrel is a full-cycle AI coding skill working across 9 AI coding agents (Claude Code, Codex, Cursor, Antigravity, Gemini CLI, GitHub Copilot, Windsurf, OpenCode, Aider) that auto-detects project state and adapts its 8-phase pipeline accordingly rather than applying a one-size-fits-all workflow. Step 0 classifies the project directory into a mode: an empty directory triggers Greenfield mode running all 8 phases from scratch, source files with no tests/docs trigger In-Progress mode (audit first, then improve), source plus tests plus CI plus README trigger Mature mode (targeted improvements only), and an explicit "fix this bug/add feature" request triggers Targeted mode (scoped work only). The 8-phase pipeline runs Discover (understand the project via code audit or requirements gathering), Plan (a concrete task list with dependencies and done-criteria), Build (write or modify code, using parallel sub-agents where the platform supports it), Test (run existing tests, write new ones, targeting 70%+ coverage), Bug Hunt (static analysis plus manual review), Polish (lint, format, type-check, remove dead code), Document (update README and inline docs without overwriting existing content), and Ship (a final checklist confirming tests are green, no secrets are present, and CI is configured). Failure recovery follows a 3-strike rule: on strike 1, fix the specific error and re-run tests; on strike 2, re-read the code and try a different approach; on strike 3, stop, revert, document what failed, and ask the user rather than continuing to force a fix. Its best practices require respecting existing code conventions (matching naming, test framework, import style, architecture, reading 2-3 similar files before writing a new one), and it explicitly never suppresses type errors with as any/@ts-ignore, never deletes failing tests to make a suite pass, and never leaves code in a broken state.

Use it when starting a new project from scratch, improving an existing codebase, fixing bugs, adding features or refactoring, adding tests/linting/CI to a project, writing production-grade documentation, or whenever the request is a multi-step "build me"/"fix this"/"squirrel this project" development task. Its limitations note it does not replace environment-specific validation or expert review, CI/CD templates it generates are starting points rather than drop-in guarantees, and parallel sub-agent execution depends on the host platform actually supporting it.

Inputs: a project directory (any state - empty, in-progress, or mature) and a development request (build, fix, improve, add tests/docs/CI).
Outputs: depending on the detected mode and phase scope - new project code built through all 8 phases, or a scoped fix/improvement applied to an existing codebase, always ending in a shippable state with green tests, no secrets, and configured CI.

npx skills add flyingsquirrel0419/squirrel-skill

Developers who want an adaptive, full-cycle coding assistant that figures out where a project actually stands and jumps in at the right phase - from greenfield builds to targeted bug fixes to hardening a mature codebase - across whichever of the 9 supported AI coding agents they use.