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/varve-sh/varve/claude-mdgit clone --depth 1 https://github.com/varve-sh/varveWrote 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/varve-sh/varve/claude-md)<a href="https://agentmods.dev/instructions/varve-sh/varve/claude-md"><img src="https://agentmods.dev/badge/instructions/varve-sh/varve/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 | $0.00898 | $0.00898 |
| Opus 5 | $0.00449 | $0.00449 |
| Sonnet 5 | $0.00180 | $0.00180 |
| Haiku 4.5 | $0.00090 | $0.00090 |
Grade A, and why
varve 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 5d 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 — 55 lines — stays where its author put it; the contents beside it link to each section on GitHub.
varve
This project has the varve MCP server connected. Use its tools for all memory operations — never use built-in memory tools or write to ~/.claude/projects/.
Memory tools
memory_pack— everything binding on a set of files, inside a token budget. The first call of a task.memory_recall— search memories by natural language query (exploration, mid-task)memory_save— save a decision, convention or note (fact/eventare accepted as synonyms fornote)memory_get— fetch full content of a memory by IDmemory_update— patch an existing note by ID (content, tags, confidence). It cannot change a memory's class: a note cannot become a decision, and an accepted decision's content is immutable.memory_forget— delete or archive a memory by ID or querymemory_context— get memories relevant to specific file pathsmemory_prompt— capture the user's goal at the start of a session
Rules (always follow)
- Before touching files — call
memory_packwith the paths you are about to read or edit and a one-line task. It returns what binds those files, deduplicated, within a budget. Pack first; recall is for questions that come up afterwards. - Before every task — call
memory_recallwith a relevant query, no exceptions. This includes commits, quick fixes, and one-liners. - Before committing — call
memory_recallto check for commit conventions. - Learn something new — call
memory_saveto persist it. - User says forget/delete/remove — call
memory_forget. - Never write memory files manually or use built-in memory features.
What actually happens when you call these tools
memory_savewithtype=decisionorconventioncreates a row with statusproposed. It does not bind, andmemory_contextwill not return it as context. A human runsvarve decision accept <id>to make it binding. Say the proposal is waiting; do not report it as adopted.memory_savewithtype=factoreventcreates a note —factandeventare synonyms fornote: retrievable, ungoverned, no lifecycle.memory_updatecannot turn a note into a decision;varve decision promote <note-id>does, and it is a human action.memory_forgeton a note deletes it. On a decision or convention it deletes nothing and changes no status: it records a disposal request and returns it as pending. A human confirms withvarve decision reject <id>while the decision is proposed, orvarve decision revert <id>once it is binding. Tell the user the request is waiting for them.memory_recallandmemory_getreturn proposals, markedPROPOSED. They are the review surface — treat anything so marked as a pending proposal, not as law.memory_contextnever returns a proposal as content; it reports them as a trailing count with their ids. Everything else it returns is binding or ungoverned.memory_packis budget-governed. It serves the top-ranked binding decisions in full, elides bodies ([body elided — memory_get <id>]) when the budget runs short, and names everything omitted in the footer with its rank and cost. Nothing is dropped silently: if it is not in the body it is in the footer. Proposals appear only as the footer count.topic_keybehaves differently by class: re-saving a note under an existing key updates it in place; re-saving a decision creates a new proposed successor that supersedes the current holder once accepted. The two namespaces are separate — a note and a decision may hold the same key.
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.
- 5d ago First seen · 55 lines · 898 tokens per session scan A 1e308fb8f1ee
varve CLAUDE.md is an instructions file published in the GitHub repository varve-sh/varve (8 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 898 tokens to every session, about $0.0045 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
ai-mind-map copilot-instructions.md
Copilot instructions for shdra06/ai-mind-map, covering ai mind map mcp — code memory engine (v1.4.0), 🚀 first thing to do in every conversation, quick lookup: "i need to..." → use this tool, ⚡ session lifecycle (always use these) and 🔍 find code (instead of grep/reading files).
THOR-memory-bible-for-AI AGENTS.md
AGENTS.md instructions for nworks3d/THOR-memory-bible-for-AI, covering agents.md - read this if you are the ai assistant, setting them up, the tools you have, how to write a note that actually comes back and after setup, this happens without you.
engineering-memlog CLAUDE.md
Claude Code instructions for atazifor/engineering-memlog, covering engineering-memlog plugin — agent instructions, how the log gets into your context, search the log yourself when warranted, append a lesson after you work and plugin knobs (env vars).
leptin AGENTS.md
AGENTS.md instructions for lionellau/leptin, covering leptin — for the agent, when to install yourself, install yourself (one block, unattended), how to use it (every session) and what to tell your human.
ai-mind-map CLAUDE.md
Claude Code instructions for shdra06/ai-mind-map, covering ai mind map mcp — code memory engine (v1.4.0), 🚀 first thing to do in every conversation, quick lookup: "i need to..." → use this tool, ⚡ session lifecycle (always use these) and 🔍 find code (instead of grep/reading files).
plur CLAUDE.md
Claude Code instructions for plur-ai/plur, covering claude.md, what is plur, development, package dependency and version bumps.