# Pilot Shell

A context and harness engineering system for Claude Code and OpenAI Codex that enforces quality gates, keeps durable project context across sessions and compaction, and turns completion claims into verified evidence.

Pilot Shell coordinates the engineering process around Claude Code and Codex rather than adding just another set of prompted rules. Hooks, stop guards, independent reviews, and full test and build gates run at every layer to prevent handoffs that merely look finished, while runtime verification, covering tests, builds, real CLI and API execution, browser automation and device checks, turns a completion claim into evidence.

Relevant source, architecture, project standards, prior decisions and session state persist across long work sessions and context compaction, with useful findings remembered automatically across both Claude Code and Codex. A companion Console brings together plan and diff review, annotations, progress tracking, evidence, session recovery and shared project knowledge, and structured slash commands add durable artefacts and explicit lifecycle contracts for specification, build, fix and PRD workflows without becoming mandatory routing for ordinary requests.

It installs with a shell script on macOS, Linux and Windows via WSL2, integrates a toolchain including ast-grep, language servers and browser automation for code discovery and impact analysis, and is written mainly in JavaScript.

Website: https://pilot-shell.com