菜单
Audit token overhead across .claude/agents/, .claude/skills/, .mcp.json, .claude/rules/, CLAUDE.md, MEMORY.md, CONTEXT.md. Identify bloat, redundancy, optimization targets. Run after adding components or when context feels heavy.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/pnils08/GodWorld --skill context-budget命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
Close a GodWorld external-lane session (Grok) cleanly: inventory work, land authorized commits if needed, rewrite only NEXT[grok] in SESSION_CONTEXT.md, commit that handoff path-specifically, report git stack / push status. Use when the user says session end, close session, handoff, wrap up, done for now, or runs /session-end. Not the Claude Code session-end skill (no PIN bump, no ROLLOUT sweep, no sessionEndMechanical.js).
Deputy Mayor Brenda Okoro's office. Generates operational statements on West Oakland Stabilization Fund, Community Development, and Economic Development (Osei-portfolio coverage). Carries Mayor's-office operational reality — not political framing.
Oakland Police Chief Rafael Montez. Generates public safety statements, OARI implementation responses, and crime data communications. Professional, measured, data-driven.
| name | context-budget |
| description | Audit token overhead across .claude/agents/, .claude/skills/, .mcp.json, .claude/rules/, CLAUDE.md, MEMORY.md, CONTEXT.md. Identify bloat, redundancy, optimization targets. Run after adding components or when context feels heavy. |
| version | 1.0 |
| updated | "2026-04-29T00:00:00.000Z" |
| tags | ["meta","infrastructure","active"] |
| effort | medium |
| sources | ["https://github.com/affaan-m/everything-claude-code/blob/main/skills/context-budget/SKILL.md (MIT, Affaan Mustafa 2026)"] |
| related_skills | ["doc-audit","tech-debt-audit","skill-audit"] |
Surface the token cost of every loaded component and rank what to trim. Token burn surfaces every few sessions (S141 Gemini autodream switch, S156 briefing-bloat audit, S180 Discord-bot edition currency). This skill makes the audit routine, not crisis-driven.
Distinct from /doc-audit (content staleness) and /tech-debt-audit (engine code health). This is purely about token economics.
.claude/agents/<name>/)Each canon-fidelity agent carries four files: IDENTITY.md + LENS.md + RULES.md + SKILL.md. Token cost per agent = sum across all four. Description-field cost is special — loaded into every Task tool spawn even if the agent is never invoked.
Surface: 25 canon-fidelity agents × 4 = 100 files. Plus engine-validator (code-only, scope-noted N/A).
Flag patterns:
.claude/skills/<name>/SKILL.md)Single SKILL.md per skill (some have supporting files). Description loads with skill registration. Body loads only when invoked.
Surface: ~80+ skills currently.
Flag patterns:
.mcp.json + plugin MCPs)Each tool schema costs ~500 tokens. Always loaded into context regardless of whether tool fires.
Surface: godworld (10 tools) + claude-mem + claude-in-chrome + + + + + + + + + + + .
figmagmailcalendardrivemaraslackcontext7discordplaywrightsupermemoryclaude-batchFlag patterns:
gh shell vs gh-MCP).claude/rules/*.md)Path-scoped per CLAUDE.md: identity.md (always), engine.md, newsroom.md, dashboard.md.
identity.mdFlag patterns:
| File | Always? | Notes |
|---|---|---|
| CLAUDE.md | yes | Project rules, auto-loaded |
| MEMORY.md | yes | Persistent-memory index, auto-loaded |
| CONTEXT.md | recommended at boot | Project glossary (S187 ADR-0001) |
| SCHEMA.md | on-demand | Doc conventions |
Flag patterns:
For each surface above, count files + tokens (words × 1.3 for prose, chars / 4 for code-heavy).
Capture per-component:
Bucket every component:
| Bucket | Criteria | Action |
|---|---|---|
| Always needed | Referenced from CLAUDE.md, backs an active skill, or matches active terminal scope | Keep |
| Sometimes needed | Domain-specific (one of N terminals, specific cycle phase) | Consider lazy-load or path-scope |
| Rarely needed | No active reference, overlapping content, no apparent home | Remove or archive |
Produce the budget report:
GodWorld Context Budget Report
═══════════════════════════════════════
Always-loaded baseline: ~XX,XXX tokens
Per-terminal additional (Full): ~X,XXX tokens
Per-cycle additional: ~XX,XXX tokens
Component Breakdown:
┌──────────────────┬────────┬───────────┬─────────────┐
│ Component │ Count │ Tokens │ Always? │
├──────────────────┼────────┼───────────┼─────────────┤
│ CLAUDE.md │ 1 │ ~X,XXX │ yes │
│ MEMORY.md │ 1 │ ~XX,XXX │ yes │
│ identity.md │ 1 │ ~X,XXX │ yes │
│ Agent descs │ 25 │ ~X,XXX │ yes (Task) │
│ Skill descs │ 80+ │ ~X,XXX │ yes (reg) │
│ MCP schemas │ N tools│ ~XX,XXX │ yes │
│ Path-scoped rules│ 4 │ ~X,XXX │ on-path │
└──────────────────┴────────┴───────────┴─────────────┘
Issues found (N):
[ranked by token savings]
Top 3 optimizations:
1. [action] → save ~X,XXX tokens
2. [action] → save ~X,XXX tokens
3. [action] → save ~X,XXX tokens
Potential savings: ~XX,XXX tokens (XX% of current overhead)
Verbose mode: per-file token counts, per-agent breakdown across IDENTITY+LENS+RULES+SKILL, full MCP tool list with per-tool schema size.
Basic audit
User: /context-budget
Skill: Scans setup → 25 agents (~30K agent files, ~12K Task-context descriptions),
80+ skills (~12K registration descriptions), 14 MCP servers w/ ~80 tools (~40K),
CLAUDE.md (~5K), MEMORY.md (~24K, at threshold).
Flags: 3 agents with descriptions > 30 words; 2 MCP servers wrapping git/gh;
MEMORY.md index entries too long (per S147 wiki-not-recall rule).
Top saving: trim MEMORY.md index → -2K; trim 3 agent descriptions → -300;
replace gh/git MCP with bash → -8K.
Verbose mode
User: /context-budget --verbose
Skill: Full report + per-file breakdown, per-agent IDENTITY+LENS+RULES+SKILL line counts,
every MCP tool with per-tool schema size estimate.
Pre-expansion check
User: I want to add a new MCP server. Do I have headroom?
Skill: Current always-loaded baseline ~85K tokens (~42% of 200K). Adding 30-tool server
= ~15K. New baseline ~50%. Recommendation: trim 2 unused MCP servers first to
stay under 45%.
words × 1.3 for prose, chars / 4 for code-heavy files.claude/rules/identity.md or a topic-specific MEMORY.md feedback file/skill-audit (skill drift) and /doc-audit (content staleness) for full health pictureaffaan-m/everything-claude-code/skills/context-budget/SKILL.md (Affaan Mustafa 2026). Localized to GodWorld component shape: 25-agent canon-fidelity stack with four files each (IDENTITY+LENS+RULES+SKILL), path-scoped rules in .claude/rules/, MEMORY.md as separate persistent layer with size threshold, CONTEXT.md as ADR-0001 vocabulary primitive, terminal-scoped persona levels (Full / Light / Stripped) factored into baseline math.