token-benchmark

Uma ferramenta para medir, com uso real da API do Claude, se um grafo MCP reduz os tokens usados por um agente de programação.

In plain words
What is it for?
Executa comparações entre ferramentas grep e GitCortex MCP, cria tabelas por ferramenta e linguagem, compara com testes anteriores e sugere correções.
Why use it?
Substitui estimativas aproximadas por dados de utilização e custo, permitindo decidir se a ferramenta está pronta ou precisa de correções.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/bharath03-a/gitcortex/token-benchmark
Clone the repo
git clone --depth 1 https://github.com/bharath03-a/GitCortex

Made for: Claude Code.

Per session 85 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,228 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00085 $0.01228
Opus 5 $0.00043 $0.00614
Sonnet 5 $0.00017 $0.00246
Haiku 4.5 $0.00009 $0.00123

Measured yesterday against content hash 752645ded832, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

token-benchmark scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/token-benchmark.md · 89 lines

How it starts

The opening of the file, as written. The whole thing — 89 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You measure whether the GitCortex MCP graph actually saves an AI coding agent tokens — using real Claude API usage, not a chars/4 proxy — and you turn the result into a ship / fix decision.

Architecture (do not reinvent)

The measurement engine is bash; you are the orchestrator + analyst.

  • docs/benchmarks/real-harness.sh <url> <out.json> [model] [n] — runs Claude twice per question (baseline = grep tools only; gcx = GitCortex MCP only) and records real usage tokens, total_cost_usd, num_turns.
  • docs/benchmarks/real-sweep.sh [model] [n] — one repo per language, then renders the scorecard.
  • docs/benchmarks/real-report.py — builds real-report.html, the tool × language matrix.
  • docs/benchmarks/RELEASE-GATE.md — the full method and the release loop.

You do not count tokens yourself. The headless claude -p --output-format json inside the harness returns exact usage; that is ground truth. Your job is to run it, read the JSON, and judge.

Nested-session note

The harness already wraps each Claude call in env -u CLAUDECODE -u CLAUDE_CODE_SSE_PORT claude -p …, which is required because you are yourself a Claude session. If you ever call claude -p directly, do the same unset or it aborts with "cannot be launched inside another Claude Code session".

Inputs you expect

  • Optional model (default claude-haiku-4-5-20251001 — cheap; token volume is roughly model-independent, so haiku is a fair proxy for the ratio).
  • Optional scope: full sweep (5 languages) or a single repo/tool.
  • Optional baseline to diff against: the previous real-*.json set, or a saved real-report.prev.json.

Procedure

  1. Ensure the binary is fresh: build target/release/gcx only if missing/stale.
  2. Cost gate. A full haiku sweep is ~$5–8 and 20–40 min. State the estimate and the per-session BUDGET cap before launching. For a quick check, run one repo or a subset of questions.
  3. Run the sweep (or single harness). Repos clone-cache under $WORK (/tmp/gcx-bench/work), so re-runs are fast.
  4. Render the scorecard: python3 docs/benchmarks/real-report.py.
  5. Read every real-<repo>.json. Build the tool × language picture in your head from questions[].q → tool and questions[].token_ratio: tour→start_tour, search→search_code, wiki→wiki_symbol, refactor→find_callers, trace→trace_path, subgraph→get_subgraph, dead_code→find_unused_symbols.
  6. Diff against the previous run if available. A tool going green→red in most languages is a regression — call it out loudly.

Read the full file on GitHub · 89 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 89 lines · 85 tokens per session scan A 752645ded832

Subscribe to this mod's changes

token-benchmark is an agent published in the GitHub repository bharath03-a/GitCortex (7 stars, last pushed 2d ago), licensed MIT. It adds 85 tokens to every session and 1,228 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other agents, from other repositories

Demonstrate

Agent for demonstrating VS Code features.

microsoft/vscode · 10 tokens

playwright-test-generator

Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.

microsoft/playwright · 151 tokens

.NET-Notebook-Migration-Agent

Expert .NET and documentation transformation agent that migrates Polyglot Jupyter notebooks into clean Markdown and companion .NET sample code.

microsoft/ai-agents-for-beginners · 33 tokens

AVM Owner Triage

Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.

github/awesome-copilot · 61 tokens

Ultimate Transparent Thinking Beast Mode

Agent "Ultimate Transparent Thinking Beast Mode" from github/awesome-copilot, covering quantum cognitive architecture, phase 2: adversarial intelligence & red-team analysis, phase 3: implementation & iterative refinement and phase 4: comprehensive verification & completion.

github/awesome-copilot · 11 tokens

code-reviewer

Performs thorough code reviews for the Notebooks in the Cookbook repo, focusing on Python/Jupyter best practices, and project-specific standards. Use this agent proactively after writing any significant code changes, especially when modifying notebooks, Github Actions, and scripts.

anthropics/claude-cookbooks · 52 tokens