Skip to content

future.md

Latest commit

 

History

History
134 lines (103 loc) · 8.64 KB

future.md

File metadata and controls

134 lines (103 loc) · 8.64 KB

Cheatcode — Post-V1 Roadmap

Everything in this file is explicitly NOT in V1. plan.md is the V1 source of truth. Items here are forward-looking only. Moving anything from this file into plan.md requires an explicit scope-extension decision.

1. Features cut from V1

These were decided out during V1 planning. They are NOT shipping in V1 and require explicit re-approval to revisit.

Feature Status Rationale
Project KB / RAG over user uploads (K1) Cut from V1 Avoids pgvector ingestion pipeline and LlamaParse runtime integration. Dedicated PDF parsing is also cut from V1 and listed below as its own future skill.
Cheatcode-as-MCP-server Cut from V1 Exposing Cheatcode as an MCP server is only valuable if developer audience traction emerges.
v0 Platform API integration Cut from V1 Vendor coupling + cost. We do UI gen via the agent + sandbox, without v0 or shadcn registry MCP tooling.
Shadcn registry MCP/tooling Cut from V1 Normal shadcn UI components stay, but no MCP package or registry-driven agent tool ships in V1.
Sensitive-tool approval gates Cut from V1 V1 does not implement generic approve/reject tool pauses. Sensitive production actions need explicit first-class commands or a future plan update.
Native iOS / Android apps Cut from V1 V1 stays web-only. Native app-store builds remain out of scope until there is clear demand.
Cheatcode Bridge (local folder sync menubar app) Cut from V1 Significant native (Tauri/Electron) effort; not justified pre-revenue.
In-sandbox always-on hosting (Zo-style services) Cut from V1 Different product category; large infra investment.
B4 — Embedded BI tool Cut from V1 Skill-based dashboard generation (csv-analyst + Recharts SSR) covers V1 use cases. Cube/Metabase-class infra is months of work.
All multi-channel inbound (email-as-trigger, SMS-as-trigger, Slack-as-trigger) Cut from V1 apps/inbound-worker, Resend Inbound, Telnyx phone provisioning, Slack inbound — all out. V1 is web-only.
Cheatcode-owned notifications Cut from V1 No persisted notification center, /v1/notifications, notification table, polling bridge, web push, email, SMS, or Slack notification channel. Active-screen feedback uses inline UI only.
Outbound transactional email Cut from V1 No Resend integration. Clerk handles auth emails (welcome, password reset, magic link); Polar handles billing receipts.
Outbound SMS Cut from V1 No Telnyx integration. No phone numbers provisioned.
Inbound webhooks for user-facing surfaces (e.g., Capy Mail-style <user>@bot.trycheatcode.com) Cut from V1 Bot subdomain + Resend Inbound + Email Workers all unused in V1. Polar/Clerk/Composio webhooks (server-to-server, no user-facing channel) stay.
Dedicated PDF parsing skill (skills/pdf-analyze) Cut from V1 V1 keeps LlamaParse as a validated BYOK provider slot only. A runtime PDF extraction skill needs a separate scope decision and eval set.
Dedicated LaTeX document skill (skills/latex-doc) Cut from V1 V1 document generation is DOCX/XLSX/PPTX/PDF through packages/tools-docs; LaTeX-specific workflows can be added later if users ask.
Music generation (Suno-style) Cut from V1 No music vendor is in the locked dependency/provider set.
Avatar video generation (HeyGen-style) Cut from V1 No avatar-video vendor is in the locked dependency/provider set.

2. v1.5 candidates (first 90 days post-V1 launch)

In rough priority order. Each requires its own scope decision before implementation begins.

  1. User-uploadable skills

    • CLI: cheatcode skills add <repo>
    • Web UI: skill marketplace browse + install
    • Validation pipeline (frontmatter schema check, eval-run before publish)
    • Trigger: V1 launch + 30 days of stable curated-skill marketplace

  2. v0 Platform API as UI generation sub-tool

    • App-builder mode delegates UI component generation to v0 for higher quality
    • Cost: $3/M tokens (v0-Pro)
    • Trigger: Lovable/v0 directly competing for our app-builder users on UI quality

  3. Connect-My-Mac (HappyCapy-style)

    • Native macOS menubar app bridging cloud agent to local Mac
    • Use case: agent generates iOS app in CF Container → triggers eas build on user's local Xcode → signs and uploads to TestFlight
    • Trigger: mobile-app skill in production with users hitting EAS Build wall

  4. AI phone calls (J7) — Vapi/Retell integration as a tool

    • Outbound: agent makes calls for appointment booking, customer service
    • Inbound: phone number per user, agent answers
    • $0.07–0.18/min (Retell), BYOK

  5. Visual automation builder (I7)

    • Zapier-style flow editor with agents as steps
    • Drag-and-drop workflow composition
    • 4–6 weeks of dedicated work; major UX surface area

  6. Conditional triggers (I8) — Event-based agent runs

    • "If revenue drops 10%, run analysis"
    • Requires (5) infrastructure to be in place
    • Webhook + threshold primitives only; cron-style recurring agents stay out of V2

  7. DB connector (B3) — Secure user DB introspection

    • Postgres, MySQL, MongoDB, Snowflake clients
    • User provides creds (BYOK)
    • RLS hardening; injection prevention; query sandboxing

  8. Liveblocks / multi-user collaboration on a project

    • Real-time cursors, presence, comments on the agent thread
    • Useful for teams reviewing pitch decks / research together

  9. Multi-channel inbound (cut from V1; re-evaluate post-launch)

    • Email-as-trigger (Capy Mail-style): <user>@bot.trycheatcode.com → starts agent on that user's default project → replies via email
    • SMS-as-trigger: Telnyx number per Pro user; inbound SMS starts agent run
    • Slack-as-trigger: bot user in user's Slack workspace responds to @-mentions and DMs
    • Requires: apps/inbound-worker, Resend Inbound config, Telnyx phone allocation flow (regulatory 10DLC for US numbers), Composio Slack OAuth + event subscription
    • Trigger to revisit: user demand signal (>20% of weekly active users requesting via Discord/support) OR a competitor closes the gap on us in app-builder lane while we open it in inbound channels

  10. Outbound transactional email + SMS

    • Resend for welcome / quota warning / digest / cancellation / win-back emails
    • Telnyx for SMS notifications on Pro+
    • V1: no Cheatcode-owned notification feature. Clerk handles auth-related email; Polar handles billing receipts.
    • Trigger to revisit: churn analysis shows users miss out-of-app notifications, or users repeatedly ask for email/SMS alerts

  11. Sandbox egress filtering proxy

    • V1 leaves Blaxel sandbox egress open; abuse is contained by sandbox resource caps, behavioral detection, and audit logging (plan §9.6)
    • v1.5: route sandbox outbound HTTP through a filtering Worker that enforces a domain allowlist (github.com, npm, pypi, common APIs) and blocks known C2/mining domains
    • Trigger to revisit: first observed egress-abuse incident, or a security-conscious customer requires it

3. v2 deferred items

Long-horizon — beyond the next 90 days. Listed here for awareness only.

  • Cheatcode Bridge — local folder sync menubar app
  • In-sandbox always-on hosting — Zo-style services where the user's sandbox runs Postgres / n8n / Discord bots persistently
  • Native iOS / Android apps — if web-only mobile usage proves insufficient
  • Self-hosted Cheatcode for enterprise customers (own Cloudflare account + own Supabase)
  • Enterprise SSO — Clerk Enterprise SAML
  • Compliance certifications — SOC 2 Type II, HIPAA BAA

4. Activation triggers

Items waiting for a specific signal before reconsidering:

Item Signal to revisit
Cheatcode-as-MCP-server Developer audience traction in V1 (>20% of users connect via Claude Code / Cursor)
Embedded BI tool 10+ users requesting it within 30 days of launch
User-uploadable skills V1 launch stable + 30 days of curated marketplace working
v0 Platform API Direct UI-quality complaints from app-builder users
Native mobile apps Repeated mobile users request native-only capabilities
Cheatcode Bridge Mobile-app skill hitting EAS Build / signing wall regularly

5. Scope discipline

Anything proposed for plan.md that exists in this file requires:

  1. Explicit scope-extension decision from the team
  2. Updated V1 acceptance criteria (what existing V1 feature gets cut to make room)
  3. Updated 8-week build plan in plan.md §17 — additions must remove or reduce existing scope, not append

The 8-week V1 build plan is locked. Drifting from it = no V1 launch.