Pragmatic Bear

Give your coding agents a real workflow.

Fifteen small skills that walk an AI coding agent from reading your repo to proving its work. The bootstrap is short; the other fourteen load only when a task calls for them.

Install Bearpaws brandonfla/bearpaws
  1. Bootstrap bp:using-bearpaws Loaded at session start. Sets skill priority and the red flags.
  2. Onboard bp:onboarding-to-a-project Reads the manifests, README, and agent files before any change.
  3. Design bp:brainstorming Shapes the change against the conventions it found.
  4. Build bp:test-driven-development Plans, tests, and debugs with whichever skill the task needs.
  5. Prove bp:verification-before-completion Runs the checks and shows the output before saying done.

What happens
at each step.

Skills are plain Markdown with a compact structure. Read any of them before you trust it.

  1. Bootstrap

    bp:using-bearpaws loads with the session. It tells the agent to check for a matching skill before it answers, lists the excuses agents use to skip one, and sets the order: onboarding, then design, then the process skills.

  2. Onboard

    bp:onboarding-to-a-project reads the manifests, the README, CLAUDE.md or AGENTS.md, and a sample of similar files, so the agent learns your stack instead of guessing from its training data. It's skipped only for abstract questions with no codebase.

  3. Design

    bp:brainstorming works out what you actually want and designs the change against the conventions onboarding found, before any code is written.

  4. Build

    The agent picks the process skill that fits the task.

    bp:writing-plans
    Breaks a multi-step task into small phases before code changes.
    bp:executing-plans
    Works through a written plan with review checkpoints.
    bp:subagent-driven-development
    Hands independent plan tasks to subagents in the same session.
    bp:dispatching-parallel-agents
    Runs two or more independent tasks at once.
    bp:using-git-worktrees
    Isolates feature work in its own worktree.
    bp:test-driven-development
    Writes the failing test first for every feature or fix.
    bp:systematic-debugging
    Finds the root cause with evidence before proposing a fix.

  5. Prove

    Done means the checks ran and passed, with the output shown.

    bp:verification-before-completion
    Runs the real commands and reads the output before claiming success.
    bp:requesting-code-review
    Checks finished work against the plan before merging.
    bp:receiving-code-review
    Verifies review feedback instead of agreeing with it on reflex.
    bp:finishing-a-development-branch
    Chooses how to merge, open a PR, or clean up once tests pass.

    The fifteenth, bp:writing-skills, is for writing and testing your own.

Where it runs.

Claude Code and Antigravity are the primary targets. The rest are experimental, and labeled that way.

Claude Code
PrimaryPlugin install. Working.
Google Antigravity IDE
PrimaryNative plugin, skills, and subagents.
Devin for Terminal
ExperimentalPartial repo-local symlink and hook wiring.
Windsurf Cascade
ExperimentalPartial repo-local symlink and rule wiring.
Codex
ExperimentalNo maintained install flow yet.

Install.

Plain Markdown and config files. Clone it, read it, change it.

Claude Code

git clone https://github.com/brandonfla/bearpaws
claude plugin marketplace add ./bearpaws
claude plugin install bp@bearpaws

Or try it for one session: claude --plugin-dir ./bearpaws

Google Antigravity

git clone https://github.com/brandonfla/bearpaws.git
cd bearpaws
./install.sh --antigravity --global

Restart Antigravity, then type /brain and check that brainstorming is listed.

Read the skills on GitHub

MIT licensed. Bearpaws began as a fork of superpowers v5.0.7 by Jesse Vincent and contributors, and now develops on its own line.