Works with the AI tools you already use

    CClaude CodeCCursorCCodex CLIGGitHub CopilotGGemini CLI+17 more

    Agent Context Budget Doctor

    by PubsProToolkit

    1

    The symptoms are familiar: instructions ignored mid-session, quality that degrades as the conversation grows, surprise costs, and compaction that eats the plan.

    Free

    1 installSecurity scanned

    See it in action

    You say

    Audit my CLAUDE.md and list of active MCP servers. The agent is getting expensive and forgets the project roadmap after about 10 messages.

    Your agent does

    Analysis shows your 'startup rent' consumes 18k tokens. The 'aws-sdk' MCP server contributes 12k tokens via 40 unused tool definitions.

    Prescription:

    1. Move the 4-page roadmap from CLAUDE.md to roadmap.md; load it only when updating status.
    2. Disable the 'aws-sdk' server for this repo.

    About this skill

    An agent's context window is a fixed budget, and most setups spend it like they haven't looked at the bill: a memory file that grew by accretion, a dozen always-loaded skills, MCP servers whose tool definitions cost tens of thousands of tokens before the first request, hooks that log every event — and then the actual work arrives to find the room half full. The symptoms are familiar: instructions ignored mid-session, quality that degrades as the conversation grows, surprise costs, and compaction that eats the plan. Agent Context Budget Doctor treats it like a clinician. Give it your CLAUDE.md or AGENTS.md, your skill list, MCP config, and hooks, plus the symptoms, and it produces a token-budget breakdown with the hogs flagged, a diagnosis mapping each symptom to its cause, and prescriptions ordered by payoff — including the rewritten memory file when trimming is the fix, a durable-state pattern that makes compaction a non-event, and the relapse-prevention habits that keep the setup lean. The download includes three reference files: the context-budget worksheet, the symptom-to-diagnosis guide, and the trimming playbook. It audits what you provide; it does not execute your agent or measure live sessions, and token figures are estimates. Works with Claude Code, Cursor, Codex CLI, Gemini CLI, and any SKILL.md agent.

    How to install

    Drop the file into your AI Agent. Works with Claude, Cursor, ChatGPT, and 20+ more.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill passes our trust checks below.

    Security scanned

    Passed our 8-point scan before listing

    1 install

    Downloaded by developers to date

    Free forever

    No account required to browse

    Frequently Asked Questions

    Popular in AI Agents & LLM Ops

    Free