AGENT SKILL

context-budget

Audit token consumption across agents, skills, MCP servers, and rules. Identifies bloat, redundant components, and produces prioritized token-savings recommendations. Use when the context window is filling up too fast or before adding new components.

License 许可
Apache-2.0
所属仓库
coco-research/coco
最近更新
2026-11-11
Stars
★ 631
Forks
15

01 前言 Front Matter

name
context-budget
tags
[ "context", "tokens", "optimization", "audit" ]
domain
meta
version
0.1.0
supports
[ "claude-code", "cursor", "codex", "generic", "windsurf", "zed", "cline", "roo-code", "amazon-q" ]
description
Audit token consumption across agents, skills, MCP servers, and rules. Identifies bloat, redundant components, and produces prioritized token-savings recommendations. Use when the context window is filling up too fast or before adding new components.

03 SKILL.md 正文

@agents/PROMPT-DEFENSE.md

Context Budget Audit

Analyze token overhead across every loaded component in a session and surface actionable optimizations to reclaim context space.

When to Use

  • Session performance feels sluggish or output quality is degrading
  • You've recently added many skills, agents, or MCP servers
  • Planning to add more components and need to know if there's room
  • Running /context-budget command (this skill backs it)

Audit Procedure

Phase 1: Inventory

Scan all component directories and estimate token consumption:

Agents (agents/*.md)

  • Count lines and tokens per file (words × 1.3)
  • Extract description frontmatter length
  • Flag: files >200 lines (heavy), description >30 words (bloated frontmatter)

Skills (skills/*/SKILL.md)

  • Count tokens per SKILL.md
  • Flag: files >400 lines
  • Check for duplicate copies — skip identical copies to avoid double-counting

Rules (rules/**/*.md, rules/**/*.mdc)

  • Count tokens per file
  • Flag: files >100 lines
  • Detect content overlap between rule files

MCP Servers (active MCP config)

  • Count configured servers and total tool count
  • Estimate schema overhead at ~500 tokens per tool
  • Flag: servers with >20 tools

CLAUDE.md / System Prompts

  • Count tokens in CLAUDE.md chain
  • Flag: combined total >300 lines

Phase 2: Classify

Sort every component into a bucket:

Bucket Criteria Action
Always needed Referenced in CLAUDE.md, backs active command, matches project type Keep
Sometimes needed Domain-specific, not referenced in CLAUDE.md Consider on-demand activation
Rarely needed No command reference, overlapping content, no project match Remove or lazy-load

Phase 3: Report

Generate a prioritized savings report:

CONTEXT BUDGET AUDIT
====================
Total estimated tokens: {total}
Context headroom: {headroom}%

TOP SAVINGS OPPORTUNITIES:
1. {component} — {tokens} tokens ({bucket}) → {recommendation}
2. {component} — {tokens} tokens ({bucket}) → {recommendation}
...

ALWAYS NEEDED (keep):
- {component} ({tokens} tokens)

SOMETIMES NEEDED (lazy-load candidates):
- {component} ({tokens} tokens)

RARELY NEEDED (removal candidates):
- {component} ({tokens} tokens)

Token Estimation Formula

  • English text: ~1.3 tokens per word
  • Code: ~1.5 tokens per word (more punctuation/symbols)
  • YAML frontmatter: ~1.2 tokens per word
  • MCP tool schema: ~500 tokens per tool definition

Integration Points

  • Run automatically before /team:ship to verify context headroom
  • Feed results to learning system as optimization instincts
  • Cross-reference with Brain DB decisions about past context issues

04 同仓库更多 Skills

coco-research
coco-research/coco

trading-super-intelligence

Trading Super Intelligence Team. Named real-world personas across 8 cells. Built local-first (LM Studio) and validator-gated. Illustrative composites; see DISCLAIMER.md.

Skill与智能体开发
★ 631
coco-research
coco-research/coco

strategy-super-intelligence

Strategy Super Intelligence Team. Named real-world personas across 7 cells. Built local-first (LM Studio) and validator-gated. Illustrative composites; see DISCLAIMER.md.

Skill与智能体开发
★ 631
coco-research
coco-research/coco

risk-compliance-super-intelligence

Risk & Compliance (GRC) Super Intelligence Team. Named real-world personas across 7 cells. Built local-first (LM Studio) and validator-gated. Illustrative composites; see DISCLAIMER.md.

律师与法务Skill与智能体开发
★ 631
coco-research
coco-research/coco

product-design-super-intelligence

Your product + design brain trust and decision partner. Named personas across 8 cells spanning product strategy, discovery/research, growth/metrics, design foundations/usability, design leadership/craft, UX/content, design systems, and a…

产品经理设计师视觉设计资料检索与研究
★ 631
coco-research
coco-research/coco

legal-ip-super-intelligence

Your legal, IP, and regulatory brain trust. 31 named personas across 7 cells covering patent strategy, tech licensing, regulatory litigation, AI governance, open-source compliance, privacy law, and antitrust.

律师与法务Skill与智能体开发
★ 631
coco-research
coco-research/coco

healthcare-life-sciences-super-intelligence

Your healthcare and life sciences brain trust. 25 named personas across 7 cells covering drug discovery, medtech, biotech strategy, health policy, genomics, clinical trials, and healthcare economics.

Skill与智能体开发
★ 631