openexp CLAUDE.md

Development instructions for openexp, including a required memory system that stores useful context and past outcomes between tasks. They describe what to record before and after work and when a user shares information.

In plain words
What is it for?
Use them when working on openexp and when searching, saving, or evaluating persistent project memory.
Why use it?
They help the project retain decisions, lessons, and predictions instead of losing that context between tasks.

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/anthroos/openexp/claude-md
Clone the repo
git clone --depth 1 https://github.com/anthroos/openexp
Per session 799 This file is loaded in full into every session.
When invoked 799 The same file — it is already loaded in full.
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 $0.00799 $0.00799
Opus 5 $0.00400 $0.00400
Sonnet 5 $0.00160 $0.00160
Haiku 4.5 $0.00080 $0.00080

Measured yesterday against content hash 95ce8604737d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

openexp 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.

CLAUDE.md · 78 lines

How it starts

The opening of the file, as written. The whole thing — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.

OpenExp — Development Instructions

Memory Protocol (MANDATORY)

OpenExp gives Claude Code persistent memory: hybrid retrieval (vector + BM25), prediction-outcome feedback, and the experience-pack pipeline. For the system to learn from your work, follow this protocol every task:

Before starting any task:

search_memory("relevant context for this task")

Find prior experience, decisions, mistakes. Hooks do auto-recall on each message, but you MUST do a targeted search before complex tasks.

After completing a task:

add_memory("what was decided/done and why", type="decision")

Capture outcomes, not just actions. The prediction loop (below) cannot reward retrieval if the outcome was never written down.

When the user shares context:

add_memory("the context", type="fact")

Immediately. Don't wait. Every piece of context improves future retrieval.

Prediction loop (build judgment over time):

When you make a prediction or recommendation (deal outcome, approach success, client reaction):

log_prediction("prediction text", confidence=0.7, memory_ids=["ids-that-informed-this"])

Later, when the outcome is known:

log_outcome(prediction_id="pred_xxx", outcome="what happened", reward=0.8)

This is the active feedback path — verified outcomes flow back to the memories that informed them. Use for: deal predictions, strategy recommendations, client behavior forecasts, technical approach bets.

Architecture

Full reference: docs/storage-system.md for retrieval/scoring details, docs/experience-library.md for the Experience Library pipeline.

  • openexp/core/ — search, scoring, lifecycle, storage
  • openexp/ingest/ — Transcript ingest + Experience Library pipeline (chunking, topic mapping, experience extraction)
  • openexp/mcp_server.py — MCP STDIO server (5 tools: search_memory, add_memory, log_prediction, log_outcome, memory_stats)
  • openexp/cli.py — CLI (search, ingest, chunk, topics, stats, compact, experience, viz)
  • scripts/batch_label.py — Batch experience labeling across all threads
  • tests/ — 300 tests across 13 files

Read the full file on GitHub · 78 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 · 78 lines · 799 tokens per session scan A 95ce8604737d

Subscribe to this mod's changes

openexp CLAUDE.md is an instructions file published in the GitHub repository anthroos/openexp (59 stars, last pushed 3mo ago), licensed MIT. It adds 799 tokens to every session, about $0.0040 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 2026-08-30.