github-twin CLAUDE.md

github-twin CLAUDE.md is an instructions file for coding agents from ChristopherDavenport/github-twin. It costs 7,884 tokens per session, scanned C, original, MIT.

Project instructions for `github-twin`, a service that searches personal or organization-wide GitHub history. They explain its modes, layout, configuration, data sources, and retrieval approach.

In plain words
What is it for?
Maintaining the GitHub-history MCP server, including ingestion, code and review retrieval, organization support, configuration, testing, and evaluation.
Why use it?
They give coding agents the context needed to work on the project and understand whether it serves one developer or an entire GitHub organization.

Instructions file

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 instructions/christopherdavenport/github-twin/claude-md
Clone the repo
git clone --depth 1 https://github.com/ChristopherDavenport/github-twin
Per session 7,884 This file is loaded in full into every session.
When invoked 7,884 The same file — it is already loaded in full.
Security scan C 1 finding. 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 $0.07884 $0.07884
Opus 5 $0.03942 $0.03942
Sonnet 5 $0.01577 $0.01577
Haiku 4.5 $0.00788 $0.00788

Measured 2d ago against content hash a4bbb2c44352, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

github-twin CLAUDE.md scanned grade C with 1 finding 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 2d ago.

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.

Reaches for credential fileshighPrivilege escalation

SSH keys, cloud credentials, git-credentials, .npmrc, /etc/shadow: reading these is how a config file becomes a credential leak.

`~/.config/gcloud/application_default_credentials.json` and never
CLAUDE.md · 547 lines

How it starts

The opening of the file, as written. The whole thing — 547 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md

Project context for agentic readers (Claude Code, etc.). User-facing docs are in README.md and getting_started.md.

What this is

A personal RAG over GitHub history exposed as an MCP server. Two target kinds, one DB per target:

  • User mode — one person's commits + review comments. The original P1/P2 scope: "write like me" + "review like me".
  • Org mode — a whole GitHub org's files-at-HEAD + commits + reviews across all members. Adds an author_login axis for filtering and per-reviewer evaluation.

Built phase by phase: P1 retrieval → P2 distillation → O-A through O-F (target abstraction, repo discovery, file-at-HEAD ingest with process-and-purge clones, org-wide commits + reviews, distill ergonomics, scale polish) → P3 predict_review_outcome → held-out eval (gt eval).

Status

Stable. 283 tests + 7 skipped (the skips are optional sentence-transformers and faiss deps). User-mode and org-mode are both functional end-to-end. Retrieval is hybrid (BM25 + vector via RRF) by default. Code chunking is AST-aware via tree-sitter for python / scala / javascript / typescript (+ tsx) / go / rust, with a line-window fallback for unsupported languages or parser failures. Embed text is prefixed with a deterministic per-chunk header (path / symbol / node-kind / leading docstring) so vector queries can hit chunks by NL even when the body contains only identifiers — see src/github_twin/embed/prefix.py and EMBED_TEXT_VERSION in pipeline.py. Bump the version constant whenever the prefix shape changes; the next gt embed wipes vec_chunk and re-embeds.

~/.local/bin/uv run pytest -q       # expect 363 passed + 7 skipped
~/.local/bin/uv run ruff check src/ tests/
~/.local/bin/uv run ruff format --check src/ tests/
~/.local/bin/uv run mypy            # strict on src/github_twin/; tests not in scope
~/.local/bin/uv run gt stats        # live user-mode DB sanity
~/.local/bin/uv run gt eval search evals/queries/default.yaml   # retrieval dogfood

Read the full file on GitHub · 547 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. 2d ago First seen · 547 lines · 7,884 tokens per session scan C a4bbb2c44352

Subscribe to this mod's changes

github-twin CLAUDE.md is an instructions file published in the GitHub repository ChristopherDavenport/github-twin (0 stars, last pushed 2mo ago), licensed MIT. It adds 7,884 tokens to every session, about $0.0394 per session on Opus 5. A static security scan graded it C with 1 finding (reaches for credential files). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.