🦑 Open Source · Python · CLI

Squidy is aContext Engineering Framework

One command generates all the documents your agent needs to work with context, rules and full traceability.

squidy — bash
▋
Read the docs →View on PyPI →

Requires Python 3.9+ and pipx

$ diff without-squidy with-squidy

The problem Squidy solves

AI agents are powerful — but without structure, every session starts from a blank slate.

Without Squidy
  • Agent with no context: starts from scratch every session
  • Project rules scattered across ad hoc prompts
  • Architecture decisions undocumented
  • Rework due to lack of traceability
With Squidy
  • Constitution and Oracle loaded automatically on boot
  • Kanban and session context always up to date
  • ADRs documented — decisions are never lost
  • Agent operational in seconds with full context

$ squidy --features

Why use Squidy?

One command, everything ready

Run `squidy` and answer a few questions. In seconds, governance files are generated and configured for your project.

Persistent context

Constitution, Oracle and session context are read by the agent on boot. It resumes exactly where it left off, without repeated questions.

Full traceability

Integrated kanban with task IDs, decision diary and ADRs. Every change has a history and justification.

LLM agnostic

Works with Claude, GPT-4, Gemini and Cursor. Generated documents follow a universal Markdown-based format.

Clear rules and prohibitions

The Constitution defines what the agent should and must never do. No surprises, no off-spec behavior.

Zero runtime dependencies

Just Markdown and YAML files. No server, no database. Works in any git repository.

$ squidy init && tree

What Squidy generates

10 structured files so any AI agent understands the project from scratch — in any session.

my-project/10 files
├─readme-agent.mdbootBoot instructions for the agent. Read before any action.
├─AGENT.mdagentSpecific instructions for Claude, Cursor and other agents. Loaded automatically in AGENT sessions.
├─constituicao.mdrulesPrinciples, prohibitions and Definition of Done for the project.
├─oraculo.mdADRsArchitecture Decision Records (ADRs) with context, decision and consequences.
├─kanban.mdtasksTask board with sequential IDs, status and acceptance criteria.
├─emergencia.mdCritical blocker log that prevents the agent from moving forward.
├─contexto-sessao.mdsessionCurrent project state: what was done, what remains, next steps.
├─politicas.mdCode conventions, naming patterns and project style standards.
└─indice-diario.mdConsolidated index of all development diary entries.
└─2026-02.mdDiary entry: what was done, why, and the outcome.