context-receipt

Claude Code loads ~36k tokens into context before you type a single word. Memory files, instruction files, skill listings, agent listings, tool schemas — all of it spends your context window before your first prompt. This prints the itemized receipt.

npx context-receipt

 CONTEXT RECEIPT
 claude code · vault
 ──────────────────────────────────────────────
 LOADED BEFORE YOUR PROMPT             tokens*
 claude code system prompt               ~6.8k
 built-in tool schemas                   ~2.4k
 CLAUDE.md (global)                       1.5k
 rules/ (3 files, global)                  486
 CLAUDE.md (project chain)                5.5k
 MEMORY.md (auto-memory index)            3.6k
 skill listing, user (115 skills)        10.6k
 skill listing, plugins (31 skills)       2.0k
 agent listing (10 agents)                 407
 ──────────────────────────────────────────────
 TOTAL                                  ~33.5k

 DEAD WEIGHT — on disk, NOT loaded
   AGENTS.md                       17.8k chars
     not read by Claude Code unless @imported from CLAUDE.md

Why care

It's not (only) the cost — startup context is cache-priced after the first message. It's the context-window tax: every always-loaded token brings compaction closer and competes with your actual work for the model's attention. Most setups accrete this invisibly — old memory entries, skills you installed once, rules files you forgot.

And the inverse surprise: dead weight — files you think are loaded that aren't. AGENTS.md sits on disk looking authoritative; Claude Code never reads it unless you @import it from a CLAUDE.md.

Usage

npx context-receipt                  # receipt for the current directory
npx context-receipt ~/code/myproj    # receipt for another project
npx context-receipt --json           # write context-receipt.json (diffable, CI-able)
npx context-receipt --budget 30000   # exit 1 if the estimate exceeds the budget

explain — computed styles for a single file

/context tells you what's loaded now. It can't tell you which instructions apply when Claude touches a specific file:

npx context-receipt explain src/auth/login.ts

 src/auth/login.ts
 inherits 7 instruction layers (~2.0k tokens est)

 1. CLAUDE.md (global)            [1.5k]   always (startup)
 5. CLAUDE.md (project)           [15]     always (startup)
 6. @import docs/extra.md         [4]      always (startup)
 7. CLAUDE.md (src)               [17]     when Claude works in this directory

 OVERLAPPING SECTIONS — same heading in multiple layers,
 all in context at once; review for contradictions:
   "git" — CLAUDE.md, src/CLAUDE.md

It resolves the global + project CLAUDE.md chain, nested per-directory CLAUDE.md files (which load lazily), and one level of @imports — then flags headings that appear in more than one layer, since those are where contradictory rules hide. Honest semantics: Claude Code loads all applicable layers simultaneously; there is no enforced cascade, so explain reports layers and overlaps rather than pretending a precedence engine exists. --json prints the structured version.

Method, honestly

• Static scan of the same files Claude Code reads: the global and project CLAUDE.md chain (walking up from the project directory), ~/.claude/rules/, one level of @imports, the project's auto-memory MEMORY.md, skill frontmatter (name + description are listed in context at startup), and agent frontmatter.
• Token counts are estimates (chars ÷ 4). Reconcile against /context inside a live session — that's the ground truth this tool approximates without needing a session.
• The Claude Code base system prompt and built-in tool schemas can't be derived statically; they're included as constants measured via /context on v2.1.212.
• Calibration: on the author's setup the static estimate lands within ~8% of the live /context total (33.5k estimated vs 36.3k measured), on the undercount side — your real number is likely a bit worse than the receipt says.

Roadmap

• Plugin skill counting, real tokenizer, /context auto-reconciliation
• context-receipt diff between two receipts (did that PR add 4k always-loaded tokens?)
• Codex CLI, Cursor, Gemini CLI receipts

MIT