IdeaWell

Project ideas from Hacker News discussions.

Claude Code Routines

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

4 Dominant Themes fromthe HN Thread

# Theme Supporting Quote
1 Sharp drop in usage limits & token‑window shrinkage – users feel throttled after the 1 M‑token window collapses back to ~200 k. “I got down to 300 tokens per minute before hitting the rate limit, compared to almost no issues when I kept under 200 k tokens.” – slopinthebag
2 Unclear ToS around third‑party / automated calls – the policy on using claude -p in scripts, cron jobs, or bots is fuzzy and often contradictory. “The TOS says you can’t access the Services via automated or non‑human means unless you’re using an official API key or explicitly permitted… yet nothing is clarified for personal automation.” – minimaxir
3 Vendor‑lock‑in / moat‑building strategies – Anthropic’s subscription rules and “Routines” feature are seen as attempts to keep users inside the ecosystem. “What they really want is a platform‑lock‑in that makes it costly to switch to OpenAI, Codex, or any other provider.” – joshstrange
4 Perceived decline in model quality & reliability – repeated complaints that newer Opus/Sonnet versions are “dumber”, hit usage limits faster, and produce inconsistent results. “Opus 4.6 now fails on basic syntax checks that it used to handle perfectly; it’s like the model is intentionally throttling token churn.” – timacles

Quotes are taken verbatim from the discussion, with attribution as shown.

🚀 Project Ideas

ContextTrim Service

Summary

  • Automatically shrinks the active context window before hitting token caps, preserving the most relevant content.
  • Prevents unexpected usage limit breaches while keeping prompts concise.
  • Works as a transparent proxy between the user and Claude Code / other LLMs.

Details

Key Value
Target Audience Developers using Claude Code Max or Pro who hit token limits frequently
Core Feature Context‑window trimming that retains salient tokens based on relevance scoring
Tech Stack Node.js API, Redis cache, Sentence‑transformer embeddings, Docker containers
Difficulty Medium
Monetization Revenue-ready: Subscription tier (Starter $9/mo, Pro $49/mo)

Notes

  • HN users repeatedly complain about “1M token window” and hidden limits, e.g., “they must let the user trim down their context window.”
  • A lightweight trim service would let users stay under caps without manually editing prompts, addressing “You used to have what? 350k tokens… now Claude will keep sending your prompt from 30 minutes ago that's completely irrelevant.”
  • Solves the frustration expressed in “The real issue stems from the 1 Million token context window change… they didn't anticipate the load” and would prevent “dumb_sugary_rock” style nerfs that reduce effective context.

Token Optimizer Scheduler#Summary

  • Dynamically selects the smallest sufficient model for each scheduled task, conserving token budget.
  • Includes a smart retry/back‑off mechanism to keep jobs running under peak load.
  • Provides alerts when remaining token quota is low.

Details

Key Value
Target Audience Teams that use Claude Code Routines or similar schedulers for nightly jobs
Core Feature Adaptive model selection + token‑budget-aware scheduling
Tech Stack Python (FastAPI), Celery + Redis, PostgreSQL, Docker Compose
Difficulty Medium
Monetization Revenue-ready: Pay‑per‑run credits ($0.001 per token saved) + optional Enterprise plan

Notes

  • References the “Daily included routine runs” limit (“15 free routine runs”) that users find restrictive.
  • Aligns with complaints about “they reduced limits mid‑subscription” and the desire for “predictable loads” to avoid “token churn”.
  • Addresses the uncertainty around “when does extra usage kick in?” and the need for a clear token‑budget view.

Open Routines Host

Summary

  • Fully open‑source, self‑hosted replacement for Claude Code Routines.
  • Supports scheduled, webhook, and event‑driven triggers while staying within any LLM’s token limits.
  • Stores workflows as portable YAML/Markdown, enabling easy migration between providers.

Details

Key Value
Target Audience Power users, open‑source advocates, and enterprises avoiding vendor lock‑in
Core Feature Plugin architecture for GitHub, Slack, cron; local workflow definitions; token‑budget alerts
Tech Stack Go microservice, SQLite, Docker, React admin UI
Difficulty High
Monetization Hobby (open‑source core, optional paid support $15/mo per instance)

Notes

  • Echoes discussions like “They want to keep you inside their ecosystem” and “I’ll leave the moment another model works for me as well as Claude does.”
  • Provides the “structured, short, schema validated state description” that HN commenters desire to replace “long chat and action log” clutter.
  • Gives users control over “checkpointing” and avoids the opaque ToS around third‑party harnesses.

Transparent Token Dashboard

Summary

  • Central dashboard aggregating token consumption across multiple LLM APIs with real‑time cap visualizations.
  • Sends proactive alerts before hitting subscription limits and suggests cheaper alternatives.
  • Generates simple reports for budgeting and compliance.

Details

Key Value
Target Audience Developers juggling multiple AI subscriptions (Claude, OpenAI, Anthropic, etc.)
Core Feature Multi‑API usage aggregation, per‑provider cap tracking, alert system
Tech Stack React front‑end, Node.js backend, Prometheus metrics, Grafana dashboards
Difficulty Low
Monetization Revenue-ready: SaaS subscription $5/mo per project

Notes

  • Directly responds to “the ambiguity is intentional” and the need for “clear documentation” about limits.
  • Users repeatedly ask “when does extra usage kick in?” and express frustration over “confusing rule changes”.
  • Offers the transparency that “they must clarify these restrictions” and prevents surprise token‑limit breaches.

Deterministic Agent Orchestrator (DAOO)

Summary- CLI tool that defines AI agent workflows in version‑controlled markdown/YAML, executes them locally or via API, and guarantees reproducible outputs.

  • Isolates LLM calls from business logic, making pipelines immune to ToS restrictions and model nerfs.

Details

Key Value
Target Audience Engineers building reliable AI‑powered pipelines who want control over execution
Core Feature Workflow DSL with checkpointing, deterministic step isolation, multi‑model fallback
Tech Stack Rust CLI, SQLite for state, OpenAPI client, Docker for isolated runners
Difficulty High
Monetization Revenue-ready: Enterprise licensing $25/mo per seat

Notes

  • Addresses “the context looks less like a long chat and action log” critique and the call for “structured, short, schema validated state description.”
  • Mirrors HN calls for “deterministic” and “schema‑validated” state, as in “a structured, short, schema validated state description, plus a short log trace.”
  • Enables “switch between providers and it comes with you,” directly answering concerns about lock‑in and “I’ll leave the moment another model works for me as well as Claude does.”

Read Later