Skip to content

Why Foundry

Build your domain layer. Skip everything else.

Eight specialized AI agents. A catalog of sixty-plus skills spanning the delivery lifecycle. One shared knowledge corpus. Every decision cited, every behavior overridable, every step reproducible.

The E11 platform's foundation, conventions, and SDK choices are already encoded — your team writes what's unique to your product.

Install Foundry →   Read the concepts →


What changes when your team uses it

Ship the first PR on day one

Cold-start engineers don't memorize the platform. The skills query @engineering11/alloy per task — pulling the right SDK doc, the right pattern, the right convention — and cite the source in every artifact. A new hire who has never seen the platform produces platform-correct code in their first session: brief → PRD → architecture → story → committed code in one sitting.

Engineers compose like the platform's authors

Architectural decisions, SDK composition rules, multi-tenancy patterns, error-handling conventions — these aren't memorized; they're queried. And not just when someone thinks to ask: the platform thread is built into the lifecycle. The PRD maps every feature to platform capabilities, the architecture inherits the map as [ADOPTED] decisions, stories carry SDK stamps, and the dev skills pre-load each SDK's conventions before writing code — capability-driven, so a task that needs outbound HTTP gets the platform's client SDK doc even when nobody named it. Reaching for axios or a raw process.env secret fails the self-check.

Audit every decision

Every non-obvious choice carries an alloy://... citation. Tests pass before tasks check off. Code review runs five adversarial layers — blind hunter, edge-case hunter, verification gap, acceptance auditor, alloy citation auditor — as context-free subagents, then triages the findings. The alloy layer verifies the pre-load record and resolves every citation on disk, so the paper trail can't quietly rot. The output isn't a black box — it's an audit trail.

Make it yours without forking it

Everything you'd want to tune — personas, menus, templates, output paths, review layers, platform gates — is data in TOML, resolved through layered overrides in .foundry/custom/ that survive every update. Add a compliance review layer, swap the PRD template, point the knowledge layer at your own runbook corpus, add a persona to the roundtable — all without touching an installed skill file. The e11-customize skill authors the overrides conversationally. See Customization.


What's actually in the install

After npx @engineering11/e11-foundry install, your project gets:

Eight agents

Every agent is a skill — a directory your AI tool reads natively; no compiled formats, no sidecars to regenerate.

The skill catalog

Lifecycle skills grouped by phase — Analysis, Planning, Solutioning, Implementation, Scaffold — plus anytime tools: help, knowledge, refine, discovery, forge-idea, multi-perspective, the four review finders, editorial passes, shard/index, customize.

Ceremony is a dial: oneshot e11-dev-flow for tiny changes, the quick flow for small work, PRD fast paths with [ASSUMPTION]-tagged drafts for features, the full lifecycle when the work warrants it — and e11-dev-auto when you'd rather not watch: it HALTs done or blocked and writes a machine-readable result for automation callers.

The knowledge layer

A single shared skill, e11-knowledge, resolves query keys (sdk:auth, pattern:multi-tenancy, matrix:unit, list:sdks) to documents in @engineering11/alloy - E11's separately-versioned knowledge corpus. The corpus ships a manifest that declares its categories, layout, and views; the skill derives structure from that manifest, so Foundry holds zero corpus-layout knowledge. When a doc is thin or missing, the skill falls back to package source under node_modules/@engineering11/<pkg>/src/ - read the platform, never guess it. Teams extend it with their own corpora via extra_query_keys.

Every claim grounded in alloy carries an alloy://<path> citation; upstream-skill material cites under sdks-nodejs-skill://. Readers can trace any platform-touching decision back to its source.

Foundry is one piece of the E11 platform; the platform pitch lives at engineering11.com.

Four tool surfaces

Same skills, installed into each tool's native skill directory:

ToolInstall target
Claude Code.claude/skills/
Cursor / Codex.agents/skills/ (shared convention)
OpenCode.opencode/skills/

Switch tools mid-project; no re-install needed. Headless, e11 agent <name> "task" runs any installed agent through a runtime, and e11 pipeline chains them through the artifacts they write to disk.

Discipline rails

The repo that produces your install holds itself to gates: vitest suites over the skills' helper scripts, reference validation across every skill file (validate:refs), guide-coverage checks so documentation can't silently drift from the catalog (validate:guide), and a zero-warning lint/format baseline enforced pre-commit. That a Foundry release passes them is part of what "disciplined AI agents" means in practice.


How it composes

text
Your domain layer        — what you actually write
     ⇧ composes with
The platform stack       — E11's deployed services and SDKs
     ⇧ documented by
The knowledge corpus     — @engineering11/alloy
     ⇧ consumed by
Foundry                  — skills that read the corpus and produce
                           code your platform already expects

You bring the domain. Foundry brings the discipline. The platform brings the foundation. The corpus connects them.


Start somewhere

If you're…Start here
New, no idea where to beginConcepts
Ready to installInstallation
Shipping a small single-contributor changeQuick flow
Starting a fresh backend repoStart a new backend repo
Scoping the agent catalogAgent quick reference
Reshaping Foundry for your teamCustomization
Extending Foundry itselfContributing

Proprietary to Engineering11 LLC.