Show HN: My Claude quota ran out in 10 minutes, so I made a tool to find out why
TL;DR Highlight
A Claude Code skill reads local logs to explain why your usage limit ran out
Who Should Read
Developers whose Claude Code usage limit suddenly runs out with no clear reason
Core Mechanics
- tare lets Claude Code read its own logs
- Claude Code logs every request locally.
- tare is a skill that reads those logs, so you just ask in English after installing.
- No accounts, packages, or configuration needed.
- Install is a single command
- Run npx skills add kelviq/tare -g -y --copy --agent claude-code.
- Start a new session and type / to check that tare appears.
- It answers with a cause, not just numbers
- The sample answer says 99% of yesterday's usage came from a tool, not the user.
- 1,553 short sessions and 9,022 requests were spawned in a website project, up to 51 at once.
- The user's own hands-on work was 93 requests.
- It deduplicates logs to fix totals
- Claude Code's log format repeats each API response several times.
- Naive counting inflated totals by 86% on the data the tool was built against.
Community Reactions
- Experience with an always-on status line
- One commenter had Claude build a plugin showing usage and context percentages at the bottom of the CLI.
- They added a hook for usage spikes and say they haven't hit caps in weeks.
- Others shared their own custom statusline projects.
- Opinion that cache expiry is the main culprit
- One comment said 95% of cases come from a large context whose cache expired.
- Another said an open-ended defect check on a codebase dispatched five subagents and burned the quota.
- Criticism that tare doesn't answer the 'why'
- A commenter said it shows where tokens went per day and tool, but the reason is left to the user.
- Their example was a missing CLAUDE.md forcing a full repo exploration.
- Another suggested adding per-project consumption.
- Opinions on distribution and built-in support
- One comment found it odd that a Python project is distributed via npm instead of PyPI.
How to Apply
- Install tare and ask about yesterday's limit
- Run npx skills add kelviq/tare -g -y --copy --agent claude-code in a terminal.
- In a new Claude Code session, enter /tare why did I hit the limit yesterday.
- If tare doesn't show after /, follow the troubleshooting in INSTALL.md.
- Check your window with /tare window before big jobs
- Before a large refactor, ask how full the 5-hour window is.
- If it is mostly full, wait until it clears.
- Find what fills your context with /tare tools
- Check how much MCP servers, subagents, and skills add to context.
- Trim or disable the heavy ones.
- Compare weeks with /tare week after adopting /clear
- After you start using /clear between tasks, compare this week with last.
- If it didn't help, ask /tare for other causes.
Try It
# Paste into a Claude Code session (after installing tare)
/tare why did I hit the limit yesterday
# Follow-up questions to try
Which of my projects is eating my quota?
Are my MCP servers adding a lot to my context?
What one change would save me the most?
# Install command (terminal)
# npx skills add kelviq/tare -g -y --copy --agent claude-codeTerminology
Related Papers
Mercury 3: Diffusion at scale
Scaling diffusion LLM 40x gives Mercury 3: 4–8x faster than speed-tier models, similar quality
Timestral: Mistral's time-series foundation models
A 209M time-series model ranks first on three benchmarks and runs on CPU.
DeepSeek V4.1 Flash 1017 (Max)
DeepSeek V4.1 Flash 1017 Max scores 45 on the Intelligence Index at 211 tok/s and $0.30 input.
If you don't run your own LLM proxy, how do you know whats going on?
Running your own LLM proxy shows which keys, models, and cache misses drive your usage.
DeepSeek: DeepSeek v4.1 Flash 1017
DeepSeek V4.1 Flash 1017: $0.05/M input, 1M context, built for agents
DeepSeek-v4.1-Flash-1017
DeepSeek V4.1 Flash minor update lifts coding and agent benchmarks while cutting run time and output length
Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware