More screenshots
Works with the AI Agents you already use
Dash
Session usage meter and task budget planner for Claude Code subscription accounts.
$2.99
Dash
Example session with this skill installed
Migrate our whole API layer from the old REST client to the v2 SDK. Every service and every call site, and update the tests so CI is green again. Go.
- Read your context and instructions
- Compiled the dash
Before I start: this touches every service and the test suite, so it's a large task (L). I'd split it across 3 parallel agents (services, shared client wrapper, tests). I ran a budget check against your Claude Code limits first.
Session (5h): [████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░] 31% used · burn 12.6%/h (recent slope) · projected 67% at reset · no limit hit before reset · resets Mon 19:47 (in 2.8h)
Weekly (7d): [█████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░] 23% used · burn 0.3%/h (window average) · projected 46% at reset · no limit hit before reset · resets Fri 04:57 (in 3.5d)
Task plan: size L, 3 agents · Migrate the API layer to the v2 client across the repo
Legend: █ used ▒ task (fits) ▓ task (overflows) ░ free
model effort session% weekly% session bar verdict
haiku high 29.0 3.5 [█████████▒▒▒▒▒▒▒▒▒░░░░░░░░░░░░] fits
sonnet low 31.0 3.7 [█████████▒▒▒▒▒▒▒▒▒▒░░░░░░░░░░░] fits
sonnet medium 51.7 6.2 [█████████▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░░] fits
sonnet high 82.8 9.9 [█████████▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓] exceeds session by 14% - wait for reset (2.8h) or downshift
opus low 52.8 6.3 [█████████▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░░] fits
opus medium 88.0 10.6 [█████████▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓] exceeds session by 19% - wait for reset (2.8h) or downshift
opus high 140.8 16.9 [█████████▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓] exceeds session by 72% - wait for reset (2.8h) or downshift
Recommendation: sonnet @ medium, ~51.7% of session, ~6.2% of week.
Agents: 3 fit this session at that config (per-agent ~15.0%). Requested count fits.
Your options
| # | Setup | Session cost | You'd end at | Trade-off |
|---|---|---|---|---|
| 1 | Sonnet, medium, 3 agents (recommended) | ~52% | ~83% | Finishes this session with headroom for fixes |
| 2 | Haiku, high, 3 agents | ~29% | ~60% | Cheapest, but weaker on cross-file refactors |
| 3 | Opus, medium, 3 agents | ~88% | over the limit | Best quality, but it won't fit: I'd do the core services now and the rest after the reset in 2.8h |
| 4 | Wait for the reset | 0% now | Start fresh at 19:47 with a full session for Opus |
Your week is fine either way: even option 3 only takes ~11% of the weekly limit, and you're projected at 46% by Friday.
Which one should I run? I won't start until you pick. Once it's done I'll run dash done, so Dash learns what a task like this really costs and the next estimate is sharper.
Connects securely to your tools. The creator never sees your data.
About this skill
Dash provides usage monitoring and budget planning for developers using Claude Code subscription accounts. It solves the problem of hitting unexpected rate limits mid-task by tracking session and weekly usage data directly from your local credentials. The tool manages the "burn rate" of your AI session, projecting when you will hit limits and estimating the cost of upcoming work based on task complexity and model selection.
What it does
- Usage tracking monitors session and weekly limits, burn rates, and projected reset times.
- Budget planning estimates the cost of tasks (XS to XL) before execution to prevent session exhaustion.
- Resource optimization recommends specific models and effort levels to fit within available capacity.
- Auto-calibration refines task cost estimates by measuring actual usage after completion.
- Visual dashboard serves a local web interface for real-time monitoring of account status.
How it works
- Run
statusto get a snapshot of current session and weekly usage percentages. - Use
planbefore large refactors or multi-agent workflows to calculate cost per model. - Select a recommended model/effort tier based on the printed budget table.
- Execute
doneafter the task to update the internal calibration model with real burn data.
Frameworks & tools
This skill is a standalone Node.js utility (version 18+) that interfaces with Claude Code configuration directories. It accesses local credentials via .credentials.json on Linux/Windows or the macOS Keychain.
Why this beats prompting it yourself
Standard prompts cannot access your local OAuth tokens or the unofficial usage endpoints required to see real-time limits. Dash automates the math of multi-agent overhead and session projections that humans typically miscalculate.
Use cases
- Prevent workflow interruptions by checking limits before starting repo-wide refactors.
- Decide between Sonnet and Haiku based on remaining weekly budget.
- Size the number of parallel agents a session can support without hitting a wall.
- Track usage across multiple Claude Code accounts from a single dashboard.
Known limitations
Requires a Claude subscription (Pro, Max, or Team); does not work with standard API keys. Usage endpoints are undocumented and may change. Local dashboard requires port 7077.
Requires Node.js 18+. Only works with Claude Code subscription accounts, not standalone API keys. Usage tracking depends on undocumented Anthropic endpoints.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
Trust & safety
Security scan
Security scan passed 7 days ago
Requires
Permissions
Allowed Hosts
Reads the user's own Claude Code OAuth token (.credentials.json, or the macOS Keychain item read with /usr/bin/security) and sends it only to api.anthropic.com/api/oauth/usage, the same endpoint Claude Code's /usage command uses. The optional dashboard runs on 127.0.0.1 only and loads nothing from the internet. Shell use is limited to opening the browser. Only env var read: CLAUDE_CONFIG_DIR. Files are written only to Dash's own data folder. No telemetry. Full list in SKILL.md under 'Network, credentials and local access'.