Gald3r

CRASH Framework

Five primitives.
Every IDE.

gald3r is built from six kinds of building blocks. Install them once into any project and your AI coding agent gains persistent memory, adversarial code review, task tracking, and cross-repo coordination — without touching your source code.

CCommandsRRulesAAgentsSSkillsHHooks
Get startedGitHub Wiki

Quickstart

Install gald3r in any project and run your first commands in under 10 minutes. Works with Cursor, Claude Code, Gemini CLI, and 32 other tools.

uv run gald3r installFirst @g-task-addRun @g-go-code

Commands

@g-* slash commands you invoke to create tasks, run reviews, plan releases, check status, and coordinate across repos. 173 commands across 12 categories.

@g-task-add@g-go-code@g-status@g-git-commit

Skills

Reusable capability modules that commands and agents load on demand. Each skill is a SKILL.md file with structured instructions — git commits, bug tracking, web crawling, memory capture. 126 skills.

g-skl-tasksg-skl-git-commitg-skl-memoryg-skl-crawl

Agents

Specialist personas (@g-agnt-*) that own a slice of the workflow. The task manager, code reviewer, verifier, and planner each maintain their own context and constraints.

g-agnt-task-managerg-agnt-verifierg-agnt-qa-engineerg-agnt-project

Rules

Always-on guidance loaded every session — git workflow conventions, documentation standards, bug discovery gates, and enforcement catchalls. Rules are never optional.

g-rl-00-alwaysg-rl-35-bug-discoveryg-rl-34-todo-lifecycleg-rl-38-component-creation

Hooks

Lifecycle scripts that fire automatically on session start, agent complete, pre-commit, and other events to inject context, enforce gates, and capture session state.

g-hk-pre-session-traceg-hk-agent-completeg-hk-pre-commitg-hk-graph-update

How a session works

  1. 1

    Session start

    A hook injects your project's .gald3r/ context — mission, active tasks, constraints — so the agent knows exactly where things stand.

  2. 2

    Create work

    @g-task-add "Add OAuth login" files a task spec with acceptance criteria, status, and history.

  3. 3

    Implement

    @g-go-code drives implementation against the task spec. The implementing agent cannot mark work done — a separate session must verify it.

  4. 4

    Review

    A fresh independent reviewer agent checks the implementation against acceptance criteria. PASS / FAIL. No access to the implementer's reasoning.

  5. 5

    Commit

    @g-git-commit produces a structured, convention-following commit with task references and correct type prefixes.

  6. 6

    Track

    @g-status and @g-report show every task's current status, blockers, and history — across the current session and all previous ones.