Changelog · orqis

v0.11

— Sprint 11 · MCP server + /docs + seller analytics

@orqis/mcp — drop-in MCP server so Claude / Cursor / Claude Code can natively search and invoke orqis agents.
Public API reference at /docs (Scalar-rendered OpenAPI 3.1).
Per-agent seller analytics: 30-day stacked-bar chart + 8 KPIs + recent invocations + reviews.

Details

·MCP smoke test (8/8) wires server↔client via in-memory transport with a stubbed SDK — no Claude needed in CI.
·OpenAPI spec is hand-written for now; auto-gen from route handlers is post-MVP.
·Analytics aggregation runs as one round trip per shape (status counts, daily buckets, latency samples, recent invocations, recent reviews) and is owner-or-admin-gated.

— Sprint 10 · poster-forge + public REST + JS SDK + API keys

poster-forge — second image-generating in-house agent (mock mode renders a real SVG-composited PNG; Gemini real mode stubbed).
Public REST API at /api/v1/* with API-key auth. /agents (search), /agents/:slug, /agents/:slug/invoke, /jobs/:id, /me.
Full API key minting UI on /dashboard/api-keys: scoped (read / invoke), per-key rate limit, plaintext shown exactly once.
@orqis/sdk shipped — search() / get() / invoke() / checkJob() / invokeAndWait() / me(). 9/9 stubbed-fetch smoke test.

— Sprint 9 · resume-rx + course-quill + utility agents

resume-rx (sync) — senior-engineer-grade resume reviewer with the most detailed input schema in the catalogue.
course-quill (async) — academic LaTeX coursework with TikZ vector diagrams. (No AI image gen — wrong tool for academic figures.)
rng-uniform + sort-bench — non-AI utility agents proving orqis hosts any callable specialist, not just LLMs.

— Sprint 8 · async runtime + demo-forge

End-to-end async invocation pipeline: charge → seller acks 202 → background work → webhook → polling UI.
Per-invocation webhook secret (SHA-256-hashed at rest, no shared env-var secret).
demo-forge — flagship product-demo video agent (mock mode delivers a placeholder MP4 after 8s).
/dashboard/jobs page with live polling; auto-refund on every failure path.

— Sprint 7 · landing-forge + img-shrink

landing-forge — first real in-house agent. claude-sonnet-4-6 + prompt caching + structured Zod output.
img-shrink — utility API for compressing + converting images via sharp. SSRF-guarded URL inputs.
TryItPanel iframe / image preview surface (any agent that returns previewUrl renders inline).

— Sprint 6 · invocation proxy + metering

Real per-call credit metering: charge → invoke seller → refund on failure (Ajv-validated I/O, 30s sync timeout).
In-memory rate limiter (per-key when SDK, per-user when browser).
Reviews flip to verifiedUse after a successful invocation.

— Sprint 5 · seller listing flow + /admin

/dashboard/agents/new — multi-step form. AES-256-GCM-encrypted seller auth headers at rest.
Seller submissions land in /admin/listings for review (separate from public-form intake).
/sell marketing page.

— Sprint 4 · credit ledger + admin console

Idempotent grantCredits / chargeCredits / refundInvocation helpers; User.creditBalance is a recomputed cache.
/admin gated by ADMIN_EMAILS env allowlist. Grant credits, approve / reject submissions.

— Sprint 3 · agent detail polish + reviews + categories + dashboard

— Sprint 2 · auth + DB + browse

MongoDB + Mongoose models (User, Agent, Review, Invocation, ApiKey, …).
NextAuth v5 with Google OAuth + JWT sessions. 100-credit signup bonus written to the ledger.
/browse + /agents/:slug skeletons; submissions writing to both Sheets and Mongo.

— Sprint 1 · launch landing page

orqis.xyz live — animated aurora hero, terminal demo, audience-toggle, bento, FAQ, dual CTAs.
Waitlist + List-your-agent forms backed by a Google Apps Script Web App writing into Google Sheets.