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/hung12ct/culi/claude-mdgit clone --depth 1 https://github.com/hung12ct/culiWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/hung12ct/culi/claude-md)<a href="https://agentmods.dev/instructions/hung12ct/culi/claude-md"><img src="https://agentmods.dev/badge/instructions/hung12ct/culi/claude-md.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.01208 | $0.01208 |
| Opus 5 | $0.00604 | $0.00604 |
| Sonnet 5 | $0.00242 | $0.00242 |
| Haiku 4.5 | $0.00121 | $0.00121 |
Grade A, and why
culi 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 4d 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.
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 — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md — Culi
Culi (cà phê culi — the single dense peaberry bean; sibling of Phin) is a context orchestrator for Claude Code and OpenAI Codex: one canonical knowledge store of small "cards" (rules / skills / lessons / styles / patterns), token-budgeted context injection via harness hooks, on-demand depth via an MCP server, and background learning from transcripts + git history. Built on gopheragent. Pure Go, one binary, no CGO.
Product goal: kill per-repo CLAUDE.md/skill duplication and inject only the most relevant context per prompt — small, dense, concentrated. Like the bean.
Full approved design: ~/.claude/plans/right-now-for-every-zazzy-graham.md (retrieval funnel,
schemas, learning pipelines, phased roadmap).
Architecture
- One binary
culi, subcommand dispatch incmd/culi/main.go(stdlibflag, no cobra). - Files are truth:
~/.culi/knowledge/(git-init'd markdown cards with YAML frontmatter) is the source of truth for cards;~/.culi/index.dbprovides SQLite FTS5 BM25 + embeddings and also holds runtime activity. Card-search tables are rebuildable, while explicit forward migrations preserve injections, card stats, session state, and metadata across schema changes. - Push:
culi hook <event>— normalized Claude/Codex UserPromptSubmit, SessionStart, Stop, and SessionEnd adapters. - Pull:
culi mcp— stdio MCP server (modelcontextprotocol/go-sdk):search_context,expand_card,save_lesson. - Learning (
internal/learn/): Claude/Codex transcript mining, style synthesis, branch→CLAUDE.md/AGENTS.md generation, cross-branch patterns — OpenAI/Anthropic API, Codex/Claude terminal, or Ollama backends behind hard daily caps. Codex lifecycle workers also perform a throttled read-only rollout scan as a missing-transcript_pathfallback; scanner telemetry is content-free and surfaced byculi doctor. - Embeddings: local Ollama (nomic-embed-text) behind gopheragent's
tools.Embedder; every path degrades gracefully to BM25-only when Ollama is down. - Review console:
culi serve— localnet/httpserver (internal/serve/) serving an embedded vanilla-JS dashboard (embed.FS, no build step): overview, candidate review, knowledge base + search, injection activity (per-event card breakdown + repo attribution), settings/repos manager. Off the hot path, so stdlib-only (C2) does not apply. Read-mostly; card actions (down/retire/ remove/edit) and config/repos writes are git-backed and respect C4 (retire = reversible status flip; hand-authored cards never round-trip through Render).
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.
- 4d ago First seen · 71 lines · 1,208 tokens per session scan A d588020cf6bd
culi CLAUDE.md is an instructions file published in the GitHub repository hung12ct/culi (2 stars, last pushed 5d ago), licensed MIT. It adds 1,208 tokens to every session, about $0.0060 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
mosaic AGENTS.md
Instructions for hexxla/mosaic, covering project instructions, your responsibilities as an agent, hexagonal architecture, core concepts and layer responsibilities & rules.
cocoindex AGENTS.md
AGENTS.md instructions for cocoindex-io/cocoindex, covering agents.md, build and test commands, building, testing and code formatting and linting.
apm python.instructions.md
Python development guidelines.
framework AGENTS.md
AGENTS.md instructions for ai-driven-dev/framework, covering agents.md, behavior, communication, action and memory management.
Commit Message
Comprehensive Conventional Commits guidelines for openwebui-extensions.
codesage AGENTS.md
Instructions for iliaal/codesage, covering codesage, build, sanity check before pushing, crate map and search pipeline.