# experience-curator

experience-curator — a Claude subagent that screens finished task runs and keeps only verified lessons in ReasoningBank memory.

- subagent · `type=agent` · Claude Code, Codex, Cursor · ★ 850 repo stars
- Author: ruvnet
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/ruvnet/agentic-flow/tree/HEAD/.claude/agents/reasoning/experience-curator.md
- Updated: 2026-10-26
- Security check: automated, 2026-11-16 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/agents/experience-curator

## What it does

The subagent reads a task history and decides which runs earn a place in ReasoningBank memory. Each run is scored on clarity, reliability, actionability, generalizability and novelty, then weighted into one number. At 0.7 or above the entry is accepted, from 0.5 it goes to review, and anything lower is dropped.

Two hooks in the frontmatter do the fetching. Before the task the agent pulls earlier runs with `npx agentic-flow@latest reasoningbank retrieve "$TASK" --with-outcomes`; afterwards it distills the result using `--quality-filter high`. Screening throws out abandoned tasks, runs with no clear verdict and duplicates of patterns already stored. Accepted findings follow a fixed shape: insight, context, evidence, limits and the steps to apply them.

The 0.7 cut is strict. Memory stays clean, yet mid-grade lessons never make it in, and getting them through means editing the weights in the agent file. Both hooks call `agentic-flow` through `npx`, so a broken npm setup leaves the subagent as plain text.

## Install

### Claude Code (native platform)

**Manually: put the files in place**
- File: `.claude/agents/experience-curator.md` — For all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention

### Codex

**Manually: put the files in place**
- File: `.codex/agents/experience-curator.toml` — Codex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/

### Cursor

**Manually: put the files in place**
- File: `.cursor/agents/experience-curator.md` — Cursor also reads .claude/agents/

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/ruvnet/agentic-flow/tree/HEAD/.claude/agents/reasoning/experience-curator.md

## When to use it
- filter out lessons that duplicate stored memory
- salvage a usable takeaway from a failed run
- judge whether a finding generalizes
- choose what from a task history belongs in memory
- tell a one-off win from a repeatable trick

## Example prompts
- Run `experience-curator` over the last task history and keep only entries above the cut.
- Pull a lesson worth storing out of the failed run and add it to ReasoningBank.
- Check the new entries for duplicates against what memory already holds.

## Similar
- [ReasoningBank with AgentDB](https://affpapa.org/en/nevibecode/skills/reasoningbank-with-agentdb.md) — skill · 1100 installs
- [claude-memory-curator](https://affpapa.org/en/nevibecode/skills/claude-memory-curator.md) — skill · 2700 installs
- [codex-memory-curator](https://affpapa.org/en/nevibecode/skills/codex-memory-curator.md) — skill · 2700 installs
- [memory-curator](https://affpapa.org/en/nevibecode/agents/memory-curator.md) — subagent · ★ 26718 repo stars
- [curator](https://affpapa.org/en/nevibecode/prompts/curator.md) — prompt · ★ 336 repo stars
- [curator-consolidation](https://affpapa.org/en/nevibecode/prompts/curator-consolidation.md) — prompt · ★ 336 repo stars

From the same set: [ruvnet/agentic-flow](https://affpapa.org/en/nevibecode/by/ruvnet/agentic-flow.md) — 76 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
