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.
npx agentmods add instructions/nikolay-e/diffctx/claude-mdgit clone --depth 1 https://github.com/nikolay-e/diffctxWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.01220 | $0.01220 |
| Opus 5 | $0.00610 | $0.00610 |
| Sonnet 5 | $0.00244 | $0.00244 |
| Haiku 4.5 | $0.00122 | $0.00122 |
Grade A, and why
diffctx CLAUDE.md 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.
How it starts
The opening of the file, as written. The whole thing — 107 lines — stays where its author put it; the contents beside it link to each section on GitHub.
diffctx
Ultimate Goal
Maximize the speed and depth of understanding textual information — for any reader, in any scenario.
Whether the consumer is an LLM processing a context window or a human reviewing a code change, diffctx's job is the same: extract the maximum signal from a codebase and present it in the clearest, most information-dense form possible. The single lens for every trade-off is comprehension-per-token — the ratio of understanding gained to attention spent.
Two Modes of Operation
Tree Mapping Mode (diffctx .) — Filesystem-focused.
Walks the directory tree respecting hierarchical ignore patterns,
reads file contents with binary/encoding detection, and serializes
to YAML/JSON/text/Markdown. Deterministic, side-effect-free.
Diff Context Mode (diffctx . --diff) — Semantics-focused.
Analyzes a git diff to intelligently select the minimal set of
code fragments needed to understand a change. For the formal
theoretical foundation, see the research paper
(source in paper/v2/, snapshot at git tag paper-v2).
Development
Setup, test commands, and the per-case YAML-corpus CI gate
(crates/diffctx-native/tests/known_below_threshold.txt, bidirectional:
a listed case that starts passing also fails) live in
CONTRIBUTING.md. Nightly CI reruns the corpus with
DIFFCTX_YAML_IGNORE_BASELINE=1, so baseline growth/shrinkage is
tracked even when every per-commit verdict passes.
Performance-change discipline (E/Q classes)
Every change is either E-class (bit-equivalent: identical selection output on identical input) or Q-class (output-changing). Q-class changes are frozen during an evaluation cycle (calibration -> validation -> sweep) — they invalidate the calibration and force a full rerun. The shared token corpus, the per-blob token cache, and the two-pass edge cap are all E-class precedents; fragmentation or scoring changes are Q-class.
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.
- yesterday First seen · 107 lines · 1,220 tokens per session scan A 63ed7d1a7c5e
diffctx CLAUDE.md is an instructions file published in the GitHub repository nikolay-e/diffctx (3 stars, last pushed 2d ago), licensed Apache-2.0. It adds 1,220 tokens to every session, about $0.0061 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.
Other instructions, from other repositories
deepagents AGENTS.md
Instructions for langchain-ai/deepagents, covering global development guidelines for the deep agents monorepo, corridor security analysis, development workflow, suppressing ruff rules and pr conventions.
autoharness AGENTS.md
Instructions for tigerless-labs/autoharness: This repo uses Open Knowledge — collaborative markdown via MCP. .open-knowledge/config.yml (with optional /.open-knowledge/config.yml; CLI/env may override) is the path contract: content.dir is the root for relative paths; content.include lists globs that add markdown…
cookiecutter-mlops-package AGENTS.md
Instructions for fmind/cookiecutter-mlops-package, covering agents.md, project overview, setup & core commands, definition of done and conventions & idioms.
python-cheatsheet copilot-instructions.md
Copilot instructions for labex-labs/python-cheatsheet, covering python cheatsheet - ai coding assistant instructions, project overview, architecture & key patterns, content structure and routing & layout system.
martex-quant CLAUDE.md
Instructions for martex-dev/martex-quant, covering session start — read these first, project instructions, ai trading bot project - master instructions, role and project objective.
ZipAgent AGENTS.md
Instructions for JiayuXu0/ZipAgent, covering repository guidelines, project structure & module organization, build, test & development commands, coding style & naming conventions and testing guidelines.