Portable Claude Code workflow framework: 40+ slash commands for the full development lifecycle, backed by a TypeScript engine for CLI and CI use. Includes
/beadfor session-handoff across repos,/vaultfor personal knowledge management,/fleet-onboardfor drift reconciliation, and/wayfinderfor decision maps.
Part of Frame OS — an AI-native application OS built as both a live product and an engineering portfolio. This repo is the shared development infrastructure that powers all Frame OS projects.
Every repo in the Frame OS ecosystem needs the same development workflows — planning, scaffolding, validating, deploying, investigating. Instead of duplicating scripts or publishing packages, this is a portable framework that installs into any sibling repo via symlink. One install-agents.sh call gives any project the full lifecycle toolkit with zero build step and instant propagation.
- Skill directories over flat scripts — each command is self-contained with its own orchestration prompt, knowledge files, and utility scripts
- Symlink installation over package publishing — zero build step, instant propagation across all ecosystem repos (shared UI components use npm publishing with Vite alias for local dev)
- 100+ ADRs — every significant decision documented with context, alternatives considered, and rationale (see ADR-0036 for the lock-file-rebuild protocol, ADR-0037 for JSONL truncation, ADR-0043 for the AgentBead bridge, ADR-0085/ADR-0070 for vault reachability, ADR-0098 for two-track telemetry, ADR-0102 for OPM inspectability)
- Self-improvement loop —
/techdebtproposes patches to the framework itself, never production code
- 40+ slash commands — plan, scaffold, validate, deploy, investigate, harden, sweep, suggest-skills, vault, fleet-onboard, wayfinder, opm, and more
- Skill-directory architecture — each command is a self-contained directory with orchestration prompt, knowledge files, and scripts
- TypeScript CLI —
core-workflowbinary for CI/CD and terminal use - VS Code extension — run any slash command from the editor
- Cross-repo installation —
install-agents.shandsession-init.shsymlinks skills + domain knowledge into sibling repos (symlinks tracked in git to survive clone and branch-switch) - Self-improvement loop —
/techdebtrecords patterns and proposes patches to the framework itself - Architecture Decision Records — 100+ ADRs documenting every significant decision across the ecosystem
| Layer | Technology |
|---|---|
| Engine | TypeScript, Anthropic SDK |
| CLI | @core/cli (core-workflow binary) |
| Editor | VS Code extension (core.runSlashCommand) |
| Testing | Vitest |
| Build | pnpm workspaces, TypeScript compiler (tsc --noEmit CI gate) |
| I want to... | Start here |
|---|---|
| Understand what this project is and how it's structured | docs/architecture.md |
| See why decisions were made the way they were | decisions/adr/ |
| Understand current priorities | decisions/okr/2026-q3.md |
| Set up and start using commands | docs/getting-started.md |
| Find the right command for a task | docs/commands.md |
| I need... | Read |
|---|---|
| Agent context before cross-repo work | domain-knowledge/frame-os-context.md |
| Available commands and architecture | CLAUDE.md |
| Skill reference material | .claude/skills/<name>/knowledge/ |
| ADR for a specific decision | decisions/adr/ |
pnpm install && pnpm buildCommands are available immediately in Claude Code — no build step required:
/plan-feature add auth to the API
/validate
/investigate why is the LangGraph node failing
/push-all
/adr new "server-side session storage over JWT"
Install into a sibling repo:
./scripts/install-agents.sh lean-canvas
# user-scope baseline (Pocock skills + ~/.claude/CLAUDE.md):
./scripts/install-agents.sh --user-scope
# ...plus the /vault skill, the ~/selfco knowledge vault, and the opt-in vault-session hook:
./scripts/install-agents.sh --user-scope --with-selfco.claude/skills/ slash commands (skill directories with knowledge/ and scripts/; catalog includes suggested_after and layer_affinity fields)
packages/
workflows/ @core/workflows — TypeScript workflow engine
cli/ core-workflow binary
vscode-extension/ VS Code extension
domain-knowledge/ Machine context corpus (loaded by commands at runtime)
decisions/
adr/ Architecture Decision Records (ADR-0001 through ADR-0102+, including ADR-0037 JSONL truncation, ADR-0043 AgentBead bridge, ADR-0085/0070 vault reachability, ADR-0098 two-track telemetry, ADR-0102 OPM inspectability)
okr/ Objectives and Key Results
docs/ Human-readable documentation
personal-knowledge/ Career context (not tracked publicly)
/techdebt is the meta-command. Other workflows trigger it when they encounter patterns worth recording. It proposes patches to packages/workflows/**, domain-knowledge/**, or skills/** — never production code.
When a mistake is caught, write it in 3 places:
- Update or add the ADR in
decisions/adr/ - Update the relevant
knowledge/file in the affected command - Update
memory/MEMORY.md
See decisions/README.md for the full write-back protocol.
All changes go through pull requests — never push directly to main.
git checkout -b feat/my-change
# make changes
git add <files>
git commit -m "feat: description of change"
git push -u origin feat/my-change
gh pr createMIT
Part of Frame OS — an AI-native application OS.
| Repo | Description |
|---|---|
| shell | Module Federation host + frame-agent LLM gateway |
| core | Workflow framework — 40+ slash commands + TypeScript engine + skill suggestion/telemetry + /vault personal knowledge (this repo) |
| FrameBus | Event bus with ADR-0013, Playwright e2e |
| cv-builder | AI-powered resume builder with LangGraph agents |
| capture-agent | Golf-course capture agent (formerly gcgcca; renamed 2026-07-30) — TX imagery → segmentation model → HF; Pydantic + TypeScript type bridge |
| blogengine | AI blog content creation platform |
| TripPlanner | AI trip planner with 11-phase pipeline |
| core-reader | Documentation viewer for the core framework |
| lean-canvas | AI-powered lean canvas business model tool |
| gastown-pilot | Multi-agent coordination dashboard |
| seh-study | NASA SEH spaced repetition study tool |
| daily-logger | Automated daily dev blog pipeline (Vercel serverless) |
| purefoy | Roger Deakins cinematography knowledge base |
| MrPlug | Chrome extension for AI UI feedback |
| frame-ui-components | Shared component library (Carbon DS) — published to npm as @ojfbot/frame-ui-components |
| beaverGame | Terrain-to-dam sandbox loop — water-as-highway game mechanic |
| asset-foundry | AI-driven Blender asset pipeline |
| silicon-empires | Claim-system strategy game — stacked-PR development pattern |
| switchboard | FastAPI proxy — byte-for-byte SSE passthrough |
| mirrorworld | Real USGS 3DEP terrain rendered as three.js heightfield |
| virtualLight | Book-to-cinema pipeline — passage extraction + cinematography-styled video prompts |
| f1-doctrine | RAQG question layer — suggests strategist questions bound to f1-substrate calls |
| morning-cockpit | Local-first morning dashboard — beads, reading, research pods |