Building the future of competitive music. A full-stack platform where artists compete in real-time battles, fans vote and the best tracks rise to the top.

.full-stack platform with real-time battle mechanics, payment integration (Stripe), AI-powered judging (OpenAI, Anthropic, Google Gemini), and a React frontend. Features include bracket-style tournaments, fan voting with virtual currency, live leaderboards, and social sharing.

Frontend: 110+ unit tests (Vitest), 0 TypeScript errors, component library with Glassmorphism/Minimalism/Bento Grid styles Backend: Django REST Framework, 70+ API endpoints, JWT auth, Stripe webhooks, OpenAI integration

Actively developing — targeting launch in 2026.

Break your app before your users do. A skill that generates worst-case user flows, runs them in a sandbox, and produces a severity-ranked disaster report.

5-step pipeline: MAP → GENERATE → SANDBOX → EXECUTE → REPORT

8 attack dimensions: Input Boundary, State Machine Torture, Temporal/Timing, AuthN/Z Shadow, Data Integrity Cascade, Concurrent Load, External Dependency Failure, Brownfield Mining

6 safeguards: No real data, no secrets, sandboxed execution, no external calls, bounded duration, no mutation

Supports greenfield (API specs) and brownfield (production logs) modes. Built on semantic anchoring — named categories force structured LLM coverage. Produces a severity-ranked disaster report with resilience score.

Usage: clone the repo, read SKILL.md, paste the assessment prompt into any agent.

Exposes the Worst Day Ever skill as an MCP server via SEP-2640 skills extension.

Built on rmcp (Rust MCP SDK) with SEP-2640 Skills extension. Exposes the skill as a skill (via io.modelcontextprotocol/skills), a tool (worst_day_ever_assess), and resources (8 dimension files).

Two transports: stdio (cargo run --bin wde-mcp-stdio) and HTTP (cargo run --bin wde-mcp-http, 127.0.0.1:3001/mcp)

Includes Python and Rust client examples. Tests: cargo test + cargo clippy --all-targets --all-features -- -D warnings.

Route, score, review, and grow AI agents through practice — not prompt-chaining.

A developmental framework where students grow through practice, feedback, challenge, and rest.

Key architecture:

• Roles over prompts — persistent personas (Student / Teacher / Principal / Janitor) with point-gated progression (0-24 → 25-49 → 50-74 → 75+)
• Compiler before critic — execute code first, LLM-as-judge last
• Adversarial review — two-judge (CTO + COO) on every output
• Difficulty-adjusted EMA scoring for growth measurement

942 tests passing, 15 skipped. Uses orca dispatch and FirstMate for crew spawning into Orca worktrees.

A two-loop SDLC in 16 installable agent skills — epic to tickets, ticket to merged PR.

Install: npx skills add branben/school-core-skills

Outer loop (human-driven planning): grill-me → vision → nose-up → to-prd → to-spec → to-tickets

Inner loop (build): Prime → Plan → Implement → Validate → Review → PR

16 skills total: grill-me, nose-up, spawn-whymage, vision, to-prd, to-spec, to-tickets, html-diagram, html-plan, html-wireframe, html-prototype, html, design-artifact, plannotator-guide, round-based-hook, think-tank, chunked-writing, implement

Skills assembled from mattpocock/skills, kunchenguid/vision, plannotator/effective-html, and plannotator/guides.

Merged: JSON Schema 2020-12 alignment for tool schemas.

PR #895 — SEP-2106: Relax outputSchema type gate ✅ Merged

• Split validate_and_strip into input/output variants: inputSchema keeps type: "object"; outputSchema now accepts any JSON Schema root type
• Updated with_output_schema docs
• 17 new tests for primitives, arrays, compositions, unit type, Json<T>, ToolBase, cache correctness, and input rejection
• Merged in v0.24.0

One endpoint. 160+ providers. 50+ free. Connect Claude Code, Codex, Cursor, Cline & Copilot to free Claude/GPT/Gemini.

Active contributor to the OmniRoute project (5.7k+ stars). The unified AI proxy/router that makes advanced AI accessible to everyone.

Contributions:

• 🔒 #2958 — MCP Scope Enforcement Fix — Fixed critical bug where dynamic tool groups (memory, skills, plugins, compression, gamification) were blocked when scope enforcement was enabled. Reordered guard logic in evaluateToolScopes() and added inline scope support to 33 dynamic tool definitions across 5 files. Credited in v3.8.8 CHANGELOG.
• 🔌 #2959 — Notion MCP Context Source — 6 MCP tools (notion_search, notion_list_databases, notion_query_database, notion_read, notion_append_blocks, notion_get_database), dashboard UI tab, read/write scope separation, 20 unit tests
• 🔧 #10575 — make GitHub skill tools discoverable through omniroute_tool_search — Fixed critical bug where the 3 GitHub skill tools were registered but invisible to omniroute_tool_search. Corrected TOTAL_MCP_TOOL_COUNT and added missing tool entries.
• 🏗️ #10039 — combo-lane awareness + activation UX + MCP visibility (Wave 2 of #9654) — Extends per-connection virtual admission lanes: fan-out awareness, operator activation control, lane state API.
• 🐛 #10322 — fix(chat-body-admission): process-wide budget (#10110) — Fixes LRU/TTL eviction orphaning live leases from Wave 1's per-connection admission lanes.
• 📡 #9940 — fix: per-connection virtual admission lanes (#9654) — Wave 1: per-connection virtual admission lanes for agentic workloads, worst-day-ever analysis, failing-test-first.
• 📝 #3077 — feat(observability): add Obsidian context source with 24 MCP tools — Obsidian via Local REST API plugin, 24 tools scoped read:obsidian/write:obsidian, follows Notion context source pattern.
• 🆕 #15983 — feat(mcp): add 2026-07-28 spec-compliant v2 server sidecar (open) — Parallel MCP server surface implementing the 2026-07-28 spec revision: explicit server/discover RPC, resultType discriminator, cacheable list responses, _meta client identity, outputSchema.
• 🐛 #15498 — fix(pii): stop redacting code syntax as IPv6 (merged) — Fixed bare :: regex alternative in pii_sanitizer.ts that corrupted Python slices, Haskell/Scala type annotations when PII sanitization was on.

Agent terminal renderer. Build CLI engine with xterm.js and PTY orchestration.

Active contributor to the Orca project (10.3k+ stars). CLI engine powering agents like Codex, Claude Code, and OpenCode in the terminal.

Contributions:

• 🖥️ #6758 — Fix TUI display artifacts over SSH — Detects in-place rewrites on background-queued output during idle, then forces a viewport refresh on first foreground chunk. Closes stale renderer-cell bug across agent TUIs.
• 🩹 #6645 — fix-maybe-win - fix escaped windows path being converted to an absolute path by winPath
• 🐳 #6671 — Support docker run outside git directory context
• 🔧 #7862 — fix(automations): deliver Hermes prompts via native query — Hermes automations now use hermes chat --query=<prompt> --tui native startup contract; prompt transported outside shell history, normalized across wrappers.

Sync your vault between desktop and mobile over Tailscale. No cloud services required.

A standalone Obsidian community plugin for vault-to-vault sync over Tailscale. Smart frontmatter filtering, backlink-aware prioritization, and end-to-end push verification.

Features:

• 🔀 Bidirectional sync over Tailscale (no cloud)
• 🏷️ Smart frontmatter filtering (skips private/personal/draft notes)
• 🔗 Backlink-aware prioritization (connected notes sync first)
• ✅ End-to-end push verification (mtime confirmation)
• 🤝 Conflict resolution (keep local/remote/keep-both)
• 🐛 Per-category debug logging with in-plugin viewer
• 🛡️ Rate limiting (120 req/min per IP)