engsys is a full team of specialized AI agents — designer, planner, implementer, reviewer, security, architect, tester, librarian — that you install into any project with one command. The personas stay the same everywhere; you swap stack "packs," not prompts.
One split runs the whole system: who does the work, which stack they work on, and what's true about this repo — kept ruthlessly separate.
Voice, values, and decision discipline that never change from project to project. Edited once, here.
Adapting means picking the AWS pack over the Azure one — not editing a prompt. Mixed stacks just compose.
Services, toolchain, invariants — written into your CLAUDE.md. The one layer unique to you.
Not one assistant wearing hats — distinct agents with distinct jobs, voices, and standards.
Slash commands call the agents by name, so the process runs the same way every time. Review findings loop a phase back; closeout lessons loop into your next project.
Capability lives in detachable packs. Install the ones a project uses; the agnostic personas load whatever is present.
A real installer — not a model — writes the folders, faithfully, every time. The model only fills in project facts afterward, so plumbing can't be silently skipped. Five steps, start to finish.
Install the CLI from npm — or git clone the repo (you'll want it anyway, to fork packs and PR lessons back). Zero-dependency Node (≥18); adds nothing to your project, same on macOS, Windows, and Linux.
npm install -g engsysScaffold the config, then pick your cloud / iac / lang / platform / db, issue tracker, and which agents you want. (From a clone instead: cp engsys/engsys.config.example.yaml engsys.config.yaml.)
engsys initMaterializes .claude/, renders CLAUDE.md + settings, and seeds the lessons library. Already have a setup? It adopts it — merges your settings, preserves your agents — and snapshots the originals so it's reversible. (From a clone: node engsys/install install --into ..)
engsys install --into .Open the project in Claude Code and run the command. It fills the project-facts region of CLAUDE.md and folds in any imported Copilot/Cursor rules. The one model-driven step — never plumbing.
/naturalizeConfirm nothing drifted, pull engsys improvements or change packs later, or roll all the way back.
engsys verify · update · uninstall --into .A config like the left produces the tree on the right:
installmaterialize, or adopt an existing setup, into a project
updatere-render; prune deselected packs; preserve your project facts + agents
verifycheck installed files against the lockfile (drift detection)
uninstallremove everything engsys added and restore the prior setup
--dry-runpreview the full plan, write nothing · --force overwrite instead of merge
CLAUDE.md, settings, and agents instead of overwriting them, imports Copilot/Cursor config for /naturalize to fold in, and uninstall restores everything byte-for-byte. Full walkthrough in install-scenarios.md.
Hard-won lessons get mined per project, generalized, and seeded into every new one. The system compounds instead of forking.