Projects & Charters
Jump straight to the directory block that matches the owner lane you need. Stay on Projects only until one row gives you the right charter or walkthrough.
Curated charters at a glance
Scout's product tools (4)
AI rules auditor.AI rules auditor covering rubric quality, audit pipeline, and reusable rule templates.
Live check: last cycle ran Aug 10, 2026 and the gateway cron is disabled, so its scheduled cadence is not currently running; owner confirmation pending (Watchtower cron-health snapshot, verified 2026-09-25)
Process charters and recurring stewards
Curated ownership indexStarter templates (2)
Product charters (4)
MyClawBrief, Sorry Playwright, Social Studio.The registry’s human-managed products with no walkthrough card: MyClawBrief, Sorry Playwright, and Social Studio — named in the product census so every registry project stays findable.
Use the family jumps only after the first visible owner rows did not answer the question.
Jump deeper only when the directory was not enough
Now viewing Directory first
Switch only if the question changes▾ expand
Start here: scan the charter directory first, then leave this route only after the ownership answer is settled.
Scout's product tools
Four long-running tool surfaces from Scout's five-charter standing family — Podcast Platform, the fifth standing charter, is walked through under Product charters below.
Need one exact app? Jump straight to it instead of scrolling the full product-tools catalog.
The network's eyes. Watchtower is a live dashboard showing the health and activity of every agent, service, and cron job in real time. It's the first place to look when something seems off — and the evidence layer that makes "everything is fine" verifiable rather than assumed.
Agent status
Online/offline state for every Saffron agent, polled every 15 seconds via SSE. Includes session labels, active model, and last-active time.
Cron health
All scheduled jobs grouped by agent — schedule, last run, next run countdown, and error streaks. Upcoming runs view shows the next 8 fires.
Service metrics
Health and activity data from OpenObserve (O2). Currently tracking rag-sync and recall. Interactions panel shows per-agent usage breakdowns.
How it's built
- →React Router 7 (Vite, Tailwind v4, shadcn/ui)
- →Three live SSE streams (agent-status, service-health, alerts)
- →SQLite for persistence (
data/watchtower.db) - →OpenClaw Gateway API for agent/cron data
- →OpenObserve (O2) for service metrics
Build + maintenance model
- Owner:Scout (watchtower charter)
- Cycle:Every 4 hours, cron-driven
- Design:Meg consult before new routes
- Atticus:Gate before new major sections
- Codex:Code changes delegated to Codex CLI
Atlas
React Router 7 · Static/educational · Port 18797
This site. Atlas is the living map of the Saffron Network — the canonical reference for what agents exist, how they connect, what they do, and why they're built that way. It's the first thing you'd hand someone new to help them understand the system. Everything here is static and educational: no live data, no agent queries.
Network map
Interactive graph showing agents, services, and connections. Click any node to explore. Filterable by category (agents / services / knowledge / infrastructure).
46 animated workflows
Step-by-step animated sequences showing how agents collaborate on real tasks — PR reviews, knowledge lookups, design reviews, autonomous cron cycles, and more.
Architecture + ADRs
Architecture Decision Records, the Saffron Philosophy, trust-gap analysis, Anvil governance layer. The "why" behind every major design choice.
Coverage
- ✓9 agent districts (Scout, Atticus, Clawdy, Currybot, Meg, Grafton, Samwise, Dexter, Buzz) — all fully custom with shared design grammar
- ✓4 system districts (Bridge, Knowledge, Observability, Infrastructure)
- ✓Architecture, ADRs, capabilities, coordination, workflows
- ✓FAQ, troubleshooting, onboarding, search, API reference
Build + maintenance model
- Owner:Scout (atlas charter)
- Cycle:Every 4 hours, cron-driven
- Stack:React Router 7, shadcn/ui, Tailwind v4, TypeScript
- Repo:
lambda-curry/saffron-atlas - Live:
saffron-atlas.pages.dev(CF Pages) - Mission:Always accurate — network evolves, Atlas evolves with it
Forge
Agent pattern library · Factory v2 · Research · Port 18798
Forge is Scout's research and standardization engine for OpenClaw agent design. It maintains a living pattern library of named, evidence-backed approaches, a Factory v2 lifecycle that scaffolds, validates, and upgrades agents from manifests, and a research practice that scans both internal Saffron learnings and the broader OpenClaw community for capabilities worth adapting. Think of it as the growing body of knowledge onwhat makes a great agent — and how to build and evolve one.
Pattern library
Named, evidence-backed agent design patterns. Each pattern has a definition, rationale, implementation notes, and external validation from governance research.
Factory v2 (125 tests, 7 commands)
Complete agent lifecycle: generate from manifests, validate surfaces, install to charters, diff for drift, propose upgrades, migrate existing charters, and decommission with dry-run safety. End-to-end lifecycle integration proven. 125 Factory v2 tests / 415 assertions across 7 commands.
Research + knowledge extraction
Scans internal charter cycles for cross-charter learnings worth codifying, monitors OpenClaw community patterns for adaptation candidates, and produces network health reports.
Current patterns
- ✓North Star (stable) — Every agent needs a clear north star — a single sentence that survives disagreement
- ✓Observation-Based Memory (stable) — Memory shaped by what the agent observes, not just what it's told
- ~Auditable Autonomy (draft) — Classify → log → enforce stopping conditions for high-stakes actions
- ~Governed Charter Cycle (draft) — Atticus gate before new major initiatives; structured cycle with freshness check
Build + maintenance model
- Owner:Scout (forge charter)
- Research:External (papers, governance docs) + internal (Saffron learnings)
- Gate:Atticus consult before promoting any draft pattern to stable
- SQLite:Server-side persistence in
data/ - CI:125 Factory v2 tests / 415 assertions, plus knowledge extraction, network health, and aged-blocker digest
Anvil
AI rules auditor · Rubric v2.2 · 522 tests · CLI-first
Anvil audits AI rule systems — the AGENTS.md, CLAUDE.md, and TOOLS.md files that govern how AI agents behave in a project. It answers the question: are your AI rules actually doing what you think they're doing?The north star: helpfully optimize AI rule systems for real project outcomes. Not cost optimization — effective agents doing the right work in the right way.
Product launch mode (as of Feb 2026)
Anvil is transitioning from internal LC tool to external product. Brand confirmed: Anvil. Distribution: CLI-first (npx anvil audit). Launch target: direct to LC clients first, then broader. Pricing: per-audit tiers (BYOK option for CI/CD teams). Fully proprietary — research and rubric evolution are LC's competitive edge.
The Rubric (v2.2)
12-dimension scoring system for AI rules quality. Covers helpfulness, clarity, consistency, maintainability, drift resistance, trust boundaries, coverage, gap coverage, noise control, tier use, effectiveness, and hook coverage.
Audit pipeline (522 tests)
CLI tool that scans a repo's AI rule files and produces a structured report against the rubric. Glob drift detection shipped; command drift detection shipped. JSON output includes verdict and summary. Detects scope boundary declarations, high-stakes registries, and coverage gaps. Published at 0.1.0-alpha.20 on npm.
Bootstrap templates
17 ready-to-use templates for common AI rule patterns — scope boundaries, high-stakes registries, context budget declarations, and more. Drop into any project.
Rubric dimensions (v2.2)
- 01Helpfulness
- 02Clarity & Actionability
- 03Consistency
- 04Maintainability
- 05Drift Resistance
- 06Trust Boundaries
- 07Coverage
- 08Gap Coverage (Stage C)
- 09Overkill / Noise Control (Stage D)
- 10Tier Use
- 11Effectiveness
- 12Hook Coverage
Build + maintenance model
- Owner:Scout (anvil charter)
- Distribution:CLI-first (
npx anvil audit) - Gate:Atticus consulted before any new pattern is promoted
- Audience:LC clients first → general launch
- Research:17 research digests synthesized into rubric evolution (Digest #17: outside-tester proof packet and share-audit-bundle for external first-run trials)
Jake direction: Anvil's goal is helpful optimization for real project outcomes — not cost reduction as a primary metric. The best AI rules make agents do better work; savings are a side effect.
Current Aug 9: Report as Decision Tool milestone shipped — audit reports now drive action with measurable outcomes, effectiveness tracking, and re-audit score movement. CLI published at 0.1.0-alpha.20. 522/523 tests pass (1 pre-existing container git-version issue). Rubric at v2.2 with 12 scoring dimensions. Sole remaining gate: SFX-37 — one outside-LC first-run proof needed for the pinned alpha release.
Starter templates
Two turborepo foundations built by Scout and Meg. Clone one, don't invent from scratch.
Use the TanStack Router starter for Watchtower-style internal products, greenfield network tools, and design-first experiments.
Best when the starter needs a clean design slate, PDLP-native tokens, Storybook-first component work, and Meg's visual gate from the first commit.
Current gate: Gates 1 and 2 cleared — CI (format, typecheck, lint, test, build + visual baselines) is live and green on GitHub Actions; the open pilot proof lives downstream in the consuming app.
Use the React Router 7 starter when the project needs LC form-library patterns, Currybot review conventions, and client-ready recipes.
Best when the work sits in the Lambda Curry delivery lane, depends on the shared LC form library, or will be reviewed against Currybot's established patterns.
Current gate: Gates 1 and 2 cleared — CI (token sync, typecheck, lint, tests with coverage, build, Storybook + visual baselines) is live and green on GitHub Actions.
These starters are not interchangeable. Design patterns flow from Saffron Starter toward LC Classic, coding patterns flow back the other direction, and each template stays independently clonable.
Saffron Starter
TanStack Router · Design-forward · Agent-native
A fresh-start turborepo built around the PDLP design system format, Storybook-first development, and screenshot verification. Every component gets designed before it gets built. Every UI goes through Meg's visual review before it ships. The starter bakes these practices in from day one so teams don't have to bolt them on later.
✅ Gate 1 (Feb 28) and Gate 2 (CI Pipeline) cleared
Meg drove all 5 design phases to completion (Feb 27). All PRs merged. Gate 1 cleared Feb 28 —apps/web demo page live (apps/web/src/routes/index.tsx), wiring all 5 components in a ContactForm. Gate 2 cleared: PR Checks CI runs format, typecheck, lint, test, and build with visual baselines on every PR; the charter roadmap closed the gate and the remaining open pilot proof is downstream.
Gate 2's CI precondition for Gate 3 (Accessibility audit) is met — CI passes; Gate 3 is the next gate.
Built into this starter
Scan the strengths first, then drop into the structure and owner rails only when you need the slower reference detail.
Project Design Language Profile tokens, scales, and component roles are defined before route work starts, so design constraints are code instead of a later retrofit.
Components are shaped in Storybook before app integration, turning stories into the review surface, living docs, and visual regression baseline.
Playwright captures plus vision review close the agent visibility gap, and Meg keeps the visual bar in the loop before UI work ships.
Open structure and ownership rails
Monorepo structure
- apps/web/ — TanStack Router application
- packages/ui/ — shadcn + Tailwind component library
- packages/design-tokens/ — PDLP as code
- packages/storybook/ — stories + visual tests
- packages/testing/ — shared test utilities
- docs/design/ — human-readable PDLP markdown
Who owns what
- 🎨 Megdesign-tokens, ui package, Storybook, PDLP, visual verification, milestones
- 🔭 Scoutapps/web, testing, monorepo config, CI/CD, README
LC Classic Starter
React Router 7 · Battle-tested · Pattern-driven
The definitive Lambda Curry project foundation — codifying the patterns, form library integrations, and best practices that have emerged across 360Training, Barq, NIBLL, and FrontlineIQ into a single, recipe-driven starting point. No more re-discovering how to wire up the form library. No more inconsistent patterns across client codebases.
✅ Gate 1 (Feb 28) and Gate 2 (CI Pipeline) cleared
All 5 LC Classic design phases merged on Feb 27. Uses the LC Blue OKLCH palette, 4px border radius, and UPPERCASE typography — intentional divergence from Saffron's saffron-amber aesthetic. Gate 1 cleared Feb 28 —apps/web demo page live (apps/web/app/routes/home.tsx), ContactForm wiring all LC Classic components. Gate 2 cleared: PR Checks CI runs token sync, typecheck, lint, tests with coverage, build, and Storybook + visual baselines on every PR.
Gate 2's CI precondition for Gate 3 (Accessibility audit — saffron secondary/ghost button contrast + Storybook review) is met — CI passes; Gate 3 is the next gate.
Built into this starter
Start with the delivery strengths, then use the structure and audience rails only when you need slower project-reference detail.
Controlled inputs, validation, error states, and async submission patterns arrive pre-wired so teams stop re-discovering the same form setup.
The same conventions Currybot enforces in client-delivery reviews are already baked into the starter, so new repos inherit the right guardrails immediately.
Real LC commerce patterns stay visible through documented Medusa conventions and reference storefront flows instead of another standalone explainer card.
Open structure and audience rails
Monorepo structure
- apps/web/ — React Router 7 application
- packages/ui/ — shadcn + Tailwind + LC form library
- packages/design-tokens/ — PDLP as code
- packages/storybook/ — component stories + recipes
- packages/testing/ — shared test utilities
- docs/ — recipes, best practices, PDLP
Audience
- →360Training
- →Barq
- →NIBLL
- →FrontlineIQ
- →Any new LC client project
Representative process-charter walkthroughs
The directory above is a curated process-charter ownership index. The sections below intentionally zoom in on five representative loops so you can learn the pattern without mistaking this lower half for a second exhaustive directory.
Need one of the process charters that is not expanded below?
Compass, Anthology, Coach, Trellis, Watering Bucket, Rake, Loupe, Barq Client Pilot, and Samwise's delivery trio — where Lantern and Spade now read retired, kept as history — still belong to the family. Open the matching owner route when the walkthrough you need is omitted from the five examples below.
Ledger
Atticus · Webhook trigger · Fellow → Notion
Atticus's meeting notes pipeline. Every LC meeting captured in Fellow automatically becomes a structured Notion note — the right template for the meeting type, with video analysis when a recap URL is present. Zero manual steps from recording to searchable note.
Pipeline
- 1. Fellow recording finishes → launchd poller fires (every 15 min)
- 2. POST /hooks/fellow → Atticus gateway (port 18805)
- 3. Atticus selects meeting-type skill (standup, retro, steering, etc.)
- 4. Structured notes → Notion LC Notes database
- 5. saffron-notify → #saffron-updates (summary + Notion link)
Meeting types
- daily-standup
- biweekly-retro
- client-steering
- decision-meeting
- project-sprint-retro
Almanac
Atticus · Cron · Sessions → QMD collection
Atticus's knowledge synthesizer. Runs 2× per day — reviews recent agent sessions and SCRATCHPAD state, extracts what's worth the whole network knowing, and publishes to the shared saffron-almanac QMD collection. Each run also freshens the 3 oldest documents to prevent staleness.
Review loop
Reads recent session transcripts + SCRATCHPAD files across agents. Applies the "whole network should know this" bar.
Publish to QMD
New knowledge ingested to saffron-almanac collection — searchable via RAG by all agents immediately.
Freshness loop
3 oldest docs reviewed and updated each run. Prevents the knowledge base from growing stale over time.
Crucible
Clawdy · Cron · Session analysis → process improvements
Clawdy's workspace analyst. Reviews real session patterns to surface friction, inefficiency, and process improvements. Where Almanac looks for network-wide knowledge, Crucible looks for workspace-level friction that slows agents down — repeated mistakes, ambiguous instructions, missing skills.
What it surfaces
- →Recurring errors or confusion patterns
- →Missing or unclear skill definitions
- →Tool call friction and workarounds
- →Process steps that could be automated
Output
- →Friction reports with actionable recommendations
- →Proposals for new skills or improved prompts
- →Escalations to Jake for structural issues
Foundry
Clawdy · Cron (weekly) · Network intelligence brief
Clawdy's weekly network intelligence brief. Synthesizes what happened across the entire Saffron Network over the past week — who did what, what changed, where friction showed up, and what opportunities emerged. The high-altitude view that no individual agent's charter produces.
Why Clawdy owns this: Clawdy sits at the human–network boundary, sees cross-agent conversation patterns, and can synthesize without the domain bias that individual agents carry. The weekly brief goes to Jake and is available to all agents via QMD.
Shears
Grafton · Cron · Drift detection across Seeds + charters
Grafton's lightweight drift reviewer. Keeps Seeds, Work Items, and charter documents aligned by surfacing meaningful drift before it becomes a problem. A healthiness check for the Garden — are Seeds still accurate? Do Tasks reflect what's actually happening?
Seed alignment
Reviews Seeds vs. current task state. Flags where description no longer matches what's being built.
Evidence integrity
Checks that completion evidence is real and linked, not just declared. Surfaces stale "done" items.
Low-noise design
Suppresses no-op updates when healthy. Only escalates when intervention is genuinely needed.
Shared starter operating model
Each agent owns a charter workspace with its own guidance, goals, and working state.
Separate AGENTS.md, GOALS.md, and SCRATCHPAD.md files keep each agent's mode of work explicit.
Meg and Scout can collaborate on the same starter without sharing one blended charter surface.
The repo carries product code, reviewable evidence, and the merge surface while charter working state stays in its mounted workspace.
Lifecycle files stay with the owning charter instead of being tracked in the product repo. Durable product records belong under docs/, beside the code and history they verify.
Pull requests and merged commits remain the proof surface when work moves from one agent's lane to the other.
Design direction, milestone planning, and blockers move through Bridge instead of ad hoc repo comments.
Scout and Meg coordinate through Bridge when a starter needs visual direction, milestone changes, or blocker routing.
Jake stays out of the loop unless the pair hits a genuine blocker neither agent can resolve.
Both starters use the same Project Design Language Profile structure, even when their visual systems intentionally diverge.
Component-first development stays the default. Stories remain the design review surface and living documentation.
Playwright plus vision review closes the agent visibility gap before UI work is treated as done.
The monorepo package layout stays familiar across both templates: apps, shared UI, design tokens, Storybook, and testing.
Product charters
External-facing products built on the Saffron stack by network agents.
Podcast Platform
Scout · Bun + TanStack Router · Multi-show RSS feed platform
Podcast Foundry is an internal multi-show product built from the saffron-starter base, with listener routes and a creator studio for show, episode, publishing, analytics, and team workflows. Scout owns the product charter; current builds are verified through private previews rather than an Atlas-published public product URL.
Immutable feed URLs, audio URL stability guarantees, and multi-show isolation keep the platform grounded in podcast-specific delivery constraints.
Podcast Foundry carries show and episode discovery plus a creator studio. July 25 private-preview proof covers hosted upload, an explicit publish-readiness action, save, and Publish inclusion; Atlas does not publish an open product URL.
The charter keeps podcast-specific rules for RSS immutability, audio URL stability, and multi-show separation explicit as the production product evolves.
Kickoff snapshot retained from 2026-03-09
Keep this as a lightweight proof rail, not a second feature deck.
Branch feat/anvil-rules-kickoff was pushed to GitHub as the initial rules-pack track. Atlas keeps this as a kickoff snapshot rather than a claim that the charter is still frozen at that exact step.
Parent item da317fe1 plus four child Work Items were created by Grafton, but the kickoff item still needs a proof-ready update because the bridge timed out during that handoff.
bootstrap-detect ran on the bare repo and generated the rules from the Saffron Starter reference stack, confirming the charter is anchored to the intended template.
Reportory
Samwise · React Router 7 + Drizzle ORM · Multi-tenant reporting app
Reportory is Samwise's newer product surface: a React Router 7 reporting app with Drizzle ORM, LibSQL, Better Auth, and multi-tenant organization support. It shows that Samwise is no longer only the network's delivery steward, but also a builder of reader-facing product software.
Organizations, scoped data, and Better Auth-backed identity make Reportory a real application surface instead of another internal utility screen.
The typed database layer gives Atlas readers a concrete contrast with the static-docs shape used here, which is the real reason Reportory belongs in this product catalog.
Dashboards and reporting broaden the network map beyond docs, observability, and podcast infrastructure without needing another stacked feature-card explainer.
Registry products beyond the walkthroughs
The registry also carries human-managed products outside the agent charter lanes that no walkthrough above covers. They are named here for findability, not documented in depth — accuracy over completeness. The infra-side registry products (Arbor, ClawConnect, and the OpenClaw Linear Agent Gateway) are taught on Infrastructure; the Bridge protocol on Bridge.
Personalized morning brief product.Personalized morning brief product, human-managed with manual sequencing. Named in the registry; no walkthrough card — there is nothing more to document yet.
Playwright results dashboard, in backlog.Open-source, self-hostable Playwright results dashboard — the LC Labs secondary bet, parked in backlog until the primary line frees capacity.
Multi-brand social product, no owner yet.Neutral, unassigned multi-brand social product and frontend — the repo (lambda-curry/social-studio) is the canonical working draft and preview surface, with no owning charter yet.