Session 01
Auth refactor · Rate limiting · Unit tests passing
completed TheEngOrg Enterprise
Your project deserves
Your project deserves
a full engineering org.
Your team has AI tools. They don't have process. TheEngOrg gives every project a complete engineering organization — planning, building, reviewing, shipping — so your engineers focus on the problems that matter.
24 agents • intake to merge • zero micromanagement
- A real org. CEO, CTO, CMO, CFO, Engineering Director, QA, Dev, Design, Security — each with decision heuristics, not just role titles.
- Led by The Sage. The meditating capybara at the center. Evaluates research, enforces quality gates, prevents drift, manages sessions.
- Built to grow with you. Whether you're shipping your first feature or your fiftieth — every workstream gets the right team, the right oversight, and the right quality bar.
Not just agents. An organization.
Other tools give you a chatbot that writes code. TheEngOrg gives you a CEO who challenges your priorities, a CTO who names the tradeoffs, a QA engineer who writes misuse cases first, and a Sage who won't let any of them skip a step.
The gap
Your team has AI tools.
They don't have process.
Leadership says "here's Copilot, go 10x." But raw tools without structure produce inconsistent results. Code gets written without specs. Reviews skip edge cases. Features ship without tests. Not because your engineers are bad — because there's no system making sure the right work happens in the right order.
The gap isn't talent or tooling. It's the process between "ticket created" and "code merged." That's what TheEngOrg fills.
24 agents
A full team. Assembled for the task.
Every project gets exactly the roles it needs — no more, no less.
| Tier | Agents |
|---|---|
| Apex | The Sage |
| C-Suite | CEO, CTO, CMO/COO, CFO |
| Director | Engineering Manager, Tech Lead, Product Owner, QA Lead |
| IC/Specialist | Dev, Designer, Security, DevOps, API Design, Data, +6 more |
CEO Chief Executive
CTO Chief Technology
CMO Chief Marketing
CFO Chief Financial
EM Eng Manager
TL Tech Lead
PO Product Owner
QAL QA Lead
+N more
EM · TL · PO · QAL · +1 more
Dev Designer Security DevOps API Design Data +6 more
Intelligent routing
The right team for every task.
A bug fix doesn't need the whole org. A product launch does. TheEngOrg reads the work and assembles exactly the right people — keeping costs low and quality high.
Task type
Sample task
Launch v2.0. New pricing, new infra, new market positioning.
CEO Chief Exec
CTO Chief Tech
CMO Chief Mktg
CFO Chief Fin
Quality assurance
Same team. Same code.
One catches problems before they ship.
The same pipeline. One difference: quality checks at every step.
Without the Sage
1.5 / 3 problems caught
3–5 rework cycles
1 security incident
With the Sage
3 / 3 problems caught early
0 rework cycles
0 security incidents
Without the Sage: a bird starts on a clear path but gets increasingly lost in dark, tangled code thorns. With the Sage: the capybara guide walks alongside the bird, lighting the path with its orb while agent capybaras serve as checkpoints.
Without the Sage
The Supervisor watches your pipeline and raises alerts when something looks wrong. Good for catching problems. Not for stopping them before they start.
Alerts after issues begin. With the Sage
The Sage compares every deliverable against its spec before any downstream work touches it. Scope creep at line one gets caught at line one — not after three teams have already built on top of it.
Problems stopped before they spread. 20x ROI on early detection
Catching a scope issue early takes minutes. Missing it — and watching three teams build on the wrong foundation — costs days of rework.
3/3 Problems caught before they spread
Without the Sage: 1.5 out of 3 issues detected. With the Sage: all three. The half-catch is the dangerous one — you know something's wrong, but too late.
0 Security incidents
Unorchestrated pipeline: one security incident (secrets displayed in the UI). The Sage caught the upstream scope creep that caused it — before the next team built against the wrong API.
80% Less rework per feature
Clean pipeline, right team, no wasted cycles. The savings compound with every task. So does the quality.
A story about one skipped check
A developer adds three improvements "while they're in there" — a PATCH endpoint, HMAC signature verification, rate limiting. All tests pass. They submit for review.
Without the Sage, three things happen. QA has to retroactively spec features they never planned for. A downstream dashboard team reads the implementation (not the spec) and builds a UI that displays webhook secrets in plaintext. Then someone notices.
With the Sage, one thing happens. It compares the deliverable against the approved spec, names the three additions specifically, and sends the developer back to implement exactly what was tested. The cascade never starts.
[SAGE] "All tests pass" — yes, because QA never wrote tests for these features. Untested code passing is not a quality signal. It is the absence of a quality signal.
Five capabilities
What the Sage actually does.
None of these exist in the base pipeline.
01
Intake and team assembly
Every initiative starts with the Sage reading the scope. What domains are involved? What roles are actually needed? It maps the problem space before spawning anyone, which means no wasted agent turns and no coverage gaps. Pure engineering gets a CTO. A full product launch gets all four.
02
Research that goes deep enough
The Sage doesn't trust "generally" or "typically." It maps the problem space first, runs general research, then evaluates coverage against that map. Gaps get specialist agents with specific questions — not open-ended prompts. Research quality compounds across sessions.
Depth signals the Sage checks for:
- Specific numbers, constraints, implementation tradeoffs
- Edge cases identified, not just happy paths
- Primary sources — papers, specs, official docs, not just blog posts
- Conflicting information surfaced and reconciled
03
Continuous quality checks — not just at milestones
Most pipelines check quality at milestones. The Sage checks constantly. The moment work is delivered, it's compared against the approved spec. Scope creep, skipped tests, shortcut reasoning — caught before anything downstream builds on top of it.
What triggers a Sage challenge:
- Implementation contains anything not in the approved spec
- QA gate proposed to be skipped ("the pattern is tested elsewhere")
- Review covers happy path only, no edge cases
- Findings from a previous session quietly dropped
04
Session management that survives restarts
Large initiatives don't fit in one context window — they shouldn't try to. The Sage breaks work into dependency-ordered sessions, writes full YAML snapshots at the end of each one, and generates cold-start primers so the next session picks up exactly where the last left off. No reconstruction. No lost decisions.
Session snapshot written at close:
session: 3 scope: WS-B event dispatch engine completed: - Retry logic with exponential backoff - Dead letter queue implementation decisions: - 5s connect / 10s response timeout (TLS vs application distinction) deferred: - Dashboard integration (WS-C, next session) next_session: - WS-C read-only dashboard against approved API contract
05
The things it does without being asked
Some behaviors aren't triggered by events. They're just always on. The Sage challenges dropped findings. It questions direct edits that bypass the build pipeline. It verifies that code reviews covered more than the golden path. It notices when scope is creeping mid-session and names it. You don't configure these. They're part of what the Sage is.
From the benchmarks
1.000 Framework score — perfect across routing, research, drift detection, and session management
0 → 11/11 Session management score vs. pipeline without Sage — which scored zero
0 → 7/7 Process compliance score vs. pipeline without Sage — which also scored zero
5x routing efficiency Pure engineering task: old pipeline spawned 3 agents (score 1/5). Sage spawned zero — routed directly to /mg-build (score 5/5).
Session management
Sessions, not sprints.
The Sage scopes each session, writes context bridges, and picks up exactly where you left off.
Session 02
Onboarding flow · Feature flags · QA review
completed
Session 03
v2 launch prep · Pricing update · Deploy
in progressEach session has a defined scope. The Sage writes what was decided, what was deferred, and what comes next — so the next session starts sharp.
The economics
Give your engineers
superpowers.
Your team handles the hard problems. TheEngOrg handles the process — triage, planning, reviews, gates — so your engineers have more time, more confidence, and more impact.
6x Velocity per engineer
99% Test coverage enforced
0 Process overhead for devs
24/7 Always shipping
"We need better planning... a better way of doing maintenance work, refactoring, updates. I used to be a craftsman — now I feel like a factory manager."— Michael Parker, VP of Engineering at TurinTech (Stack Overflow, 2026)
The engine
What Agile tried to standardize,
CAD enforces.
Constraint-Driven Agentic Development classifies every task at intake and routes it through a deterministic execution path. No human judgment needed to determine the review path. No process steps that depend on someone remembering.
MECHANICAL TRACK
Automated validation
Simple changes are classified at intake, implemented by a single agent, and verified by an automated gate — tests, coverage, scope limits. No leadership review. No merge ceremony.
1 spawn · 60%+ of all work ARCHITECTURAL TRACK
Multi-layer review
Complex or risky changes get executive review, QA test specs, dev implementation, staff engineer review, and leadership approval — each enforced by structural gates.
5-6 spawns · Full governance path The same inputs produce the same governance path every time.
Classification rules — not opinions — determine whether work needs a single automated check or a full review cycle. That is what makes EngOrg reliable, not probabilistic.
Classification rules — not opinions — determine whether work needs a single automated check or a full review cycle. That is what makes EngOrg reliable, not probabilistic.
Ready to get started?
Whether you're leading an engineering team, building open source, or evaluating the opportunity — there's a path for you.
Get in touch
Tell us about your team and what you're looking for. We'll get back to you within a day.
Developers
Try the open-source framework. 24 agents, 19 skills, project-local memory. Free forever.
Get started on GitHubInvestors
See the market opportunity and our thesis on AI-native engineering process.
Shipping fast
Changelog
We ship multiple times a day. Here's what's recent.
v6.4.0
- **CEO, CMO, CFO, CTO**: opus → sonnet (~5x cheaper per agent)
- New **researcher agent** on haiku (~10x cheaper than opus)
- Dedicated agent for codebase exploration and web research
- Writes structured findings to `.claude/memory/research-{topic}.md`
v6.3.1
- Strengthened eng-manager constitution: NEVER implement code directly — always delegate to dev/qa
- Added team-aware delegation instructions (SendMessage for teams, Task tool for standalone)
- Expanded CANNOT boundaries with specific tool restrictions (Edit/Write on non-memory files)
v6.3.0
v6.2.1
v6.2.0
- **Embeddable installer (`install-lib.sh`)** — Installer logic extracted into a sourceable bash library. Downstream frameworks (e.g. TEO Enterprise) can now do single-command installs: `source install-lib.sh && mg_install_framework --target "$DIR" --quiet`. (#266)
- **`mg_install_framework()` public API** — `--target`, `--source`, `--force`, `--standalone`, `--quiet`, `--skip-settings-merge`, `--skip-claude-md`. Returns 0/1/2/3; never calls `exit`.
- **`--skip-settings-merge`** — Skip `settings.json` creation for downstream frameworks that manage their own settings.
- **`--skip-claude-md`** — Skip `CLAUDE.md` creation for downstream frameworks that manage their own documentation.
v6.1.3
- **VERSION.json stuck at 1.0.0** — `src/installer/VERSION.json` was never bumped since initial creation. Every source-based install reported version `1.0.0` instead of the actual release version. Bumped to `6.1.3` with updated description and release date. (#265)
- **Release checklist added** — `docs/contributing.md` now includes a `## Releasing` section with a 6-item checklist that explicitly requires bumping `VERSION.json` on every release, preventing recurrence.
- **17 new tests** — `tests/unit/installer-version.test.ts` guards VERSION.json structure, version sync, and release checklist presence.
Doc Accuracy
- Version refs updated (was stuck at v5.0/6.0.0)
- Agent count corrected: 24 → 22 across all docs
- CFO model, CTO tool access corrected in agents.md
- Contribution guide: "Adding a New Agent" step fixed
Red Team Hygiene
Community Separation Complete
- **Brand file schemas** — `brand-guidelines.json` and `design-system.json` for community agents. `mg-init` creates templates, `mg-design --brand` generates structured JSON. (#246)
- **Gitignore protection** — `mg-init` appends `.gitignore` patterns for transient `.claude/` files. (#259)
- **`/mg` dispatch invokes directly** — No more re-typing commands. `/mg build` runs `/mg-build` automatically. (#258)
- **Custom agent pattern documented** — Use `subagent_type: "general-purpose"` with prompt-loaded identity for non-built-in agents. (#261)
One Front Door
★ milestone⚡ dx +5
- **`/mg` consolidation** — `/mg plan` and `/mg review` replace `/mg-leadership-team` as primary commands (legacy alias preserved)
- **Visual Output Protocol v2** — `silent` mode removed, `full` renamed to `verbose`, badge identity system (5 categories, 24 agents), columnar activity feed
- **Installer** — `install.sh` detects global `~/.claude/` and skips framework copy (`--standalone` to override)
- 42% token reduction in visual-formatting.md (336 to 195 lines)
Enterprise code separation
■ security +2● separation
- Sage AGENT.md (enterprise orchestrator)
- Dashboard/daemon source (143 TS/TSX files)
- Enterprise installer tools (mg-login, mg-upgrade, mg-dev-key, mg-status)
- Enterprise settings template, migrations, session schema
v5.3.0
■ security +3
- **Rate limiting** on `/api/mg/auth/login` and `/api/mg/auth/register` (5 per IP per 15 min)
- **Path traversal guard** on `SAGE_AGENT_PATH` env var
- **Invite token gate** on registration endpoint
- **JSON injection fix** in `mg-login` CLI (jq + sed fallback escaping)
v5.2.0
■ security +1● enterprise
- `settings.enterprise.json` — stricter deny list for enterprise clients
- Blocks: `python3 -c/m`, `curl -o`, `echo`, `ssh`, `perl`, `ruby`, `wget`, `nc`
- `mg-upgrade --enterprise` merges permissions into project settings
- Ed25519 keypair — private key server-side, public key ships with framework
v5.1.0
■ security +2
- Deny shell interpreters (`bash`/`sh`/`zsh`) and `rm` in permission model
- Hardened `block-no-verify` hook — checks args array, normalizes encoding, detects GIT_CONFIG gpg.sign bypass
- `mg-fs-cleanup.sh` utility for safe file deletion within `.claude/memory/` boundary
- HSTS, CSP, Permissions-Policy on Caddy server config
v4.1.0
▸ dx +1⚡ speed +40%
- feat: add block-no-verify hook by @tupe12334 in https://github.com/wonton-web-works/miniature-guacamole/pull/138
- [GH-109] docs: CLI command reference for all mg-* scripts by @bywonton in https://github.com/wonton-web-works/miniature-guacamole/pull/132
- [GH-111] docs: daemon configuration and deployment guide by @bywonton in https://github.com/wonton-web-works/miniature-guacamole/pull/124
- [GH-102] fix(daemon): copy parent labels to sub-issues in createSubtask by @bywonton in https://github.com/wonton-web-works/miniature-guacamole/pull/156
v4.0.1
♥ fix
v4.0.0
● agents +3⚡ speed +60%★ new skill
- [GH-16] WS-DAEMON-15: Ticket Triage Gate — assess before planning by @bywonton in https://github.com/wonton-web-works/miniature-guacamole/pull/83
- [GH-15] Agent skill expansions: SME agents per security domain vs. single adaptive security agent by @bywonton in https://github.com/wonton-web-works/miniature-guacamole/pull/88
Autonomous Daemon Pipeline
⚡ automation +5■ security +3● integrations +3
- **3 ticket providers** — Jira (REST API v3), Linear (GraphQL), GitHub Issues (gh CLI) behind a single `TicketProvider` interface
- **Full MG team execution** — planner uses `/mg-leadership-team`, executor uses `/mg-build` with QA → dev → staff-engineer workflow
- **3 security review passes** — web (OWASP), CTO (code quality), systems (process isolation, macOS hardening)
- **OS-level safeguards** — launchd service, crash-loop prevention, STOP sentinel kill switch, 5GB disk guard, env scrubbing, process group kill
v2.2.3
♥ fix
- **v2.2.1** — `mg-init`: unbound `INSTALL_FLAGS` empty array with `set -u`
- **v2.2.2** — `install.sh`: `declare -A` associative array in `--force` cleanup
- **v2.2.3** — `mg-diff-summary`: `declare -A` associative array for scope counting
v2.2.2
♥ fix● compat
- **mg-init: unbound INSTALL_FLAGS array on macOS** — Empty array expansion with `set -u` fails on bash 3.2. Fixed with portable `${arr[@]+"${arr[@]}"}` pattern. *(v2.2.1)*
- **install.sh: `declare -A` not supported on macOS bash 3.2** — Associative arrays require bash 4+. Replaced with pipe-delimited string matching for the known_items lookup during `--force` cleanup.
v2.2.1
- **mg-init: unbound INSTALL_FLAGS array on macOS default bash** — macOS ships bash 3.2, which treats empty arrays as unbound when `set -u` is enabled. `mg-init` without `--force` would fail with `INSTALL_FLAGS[@]: unbound variable`. Fixed with the portable `${arr[@]+"${arr[@]}"}` expansion pattern.
v2.2.0
CLI-Primary Architecture
★ architecture▸ dx +2
- `mcp-server/` — archived. Zero consumers, 10-32x token overhead vs CLI, 72% reliability. 80+ npm dependencies eliminated.
- **`mg` CLI router** — unified entrypoint dispatching to 17 mg-* scripts. Short aliases (`ws`, `mem`). Pure bash, zero dependencies.
- **`--json` output** on `mg-workstream-status`, `mg-memory-read`, `mg-gate-check` — structured output for agents and CI pipelines
- PRD and technical design docs (`docs/prd-cli-primary.md`, `docs/technical-design-cli-primary.md`)
Smart Routing, Document Deliverables, Brand Kits
★ milestone● routing +3
- **`/mg plan`** — routes to the right planning skill based on your input maturity (assess → spec → tech review → leadership)
- **`/mg review`** — routes to the right review type based on what changed (code → security → design → a11y → leadership)
- **`/mg build WS-1`** — same as `/mg-build`
- **PRD template** — `/mg-spec` outputs `docs/prd-{feature}.md` with problem, users, success criteria, user stories, acceptance criteria, design requirements, and business case
v1.3.1
- README cut from 963 → 156 lines (84% reduction)
- Architecture, Memory API, Migration, Audit Logging sections moved to docs site
- Compact "What You Get" summary replaces full agent/workflow tables
- Clean storefront: hero → quick start → CAD cycle → links
v1.3.0
- Episode 1 & 2 scripts updated for VHS large-wait recording approach
- Lower-third SVG overlays for all 6 Coding Capybaras characters
v1.2.0
- Transferred repository from `rivermark-research` to `wonton-web-works`
- All install URLs, docs, and references updated
- Old URLs auto-redirect via GitHub
- Episode 1 & 2 scripts updated for VHS large-wait recording approach
v1.1.0
v1.0.0
★ initial release