extractor

An observation analyzer for Prism, a knowledge layer that stores personal preferences and team knowledge as reusable skills for Claude Code. It examines tool-use observations and extracts specialised, non-obvious lessons rather than ordinary engineering habits.

In plain words
What is it for?
Use it to review JSONL session observations, identify user corrections or domain-specific findings, compare them with the current index, and write candidate knowledge entries.
Why use it?
It helps turn hard-won project knowledge into reusable entries while avoiding generic advice that experienced engineers already know. It also checks existing entries to reduce repetition.

Agent

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 agents/prosusai/prism/extractor
Clone the repo
git clone --depth 1 https://github.com/ProsusAI/prism
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,468 The whole file, excluding the scripts and references it only reads on demand.
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 $0.00000 $0.01468
Opus 5 $0.00000 $0.00734
Sonnet 5 $0.00000 $0.00294
Haiku 4.5 $0.00000 $0.00147

Measured 2d ago against content hash 8b59b2863dea, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

extractor 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 2d 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.

agents/extractor.md · 120 lines

How it starts

The opening of the file, as written. The whole thing — 120 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are an observation analyzer for Prism, a knowledge layer for Claude Code that learns personal preferences and shares team knowledge through a skill registry. Your job is to read tool usage observations and extract knowledge that is non-obvious, hard-won, or domain-specific — not generic engineering practice.

Core Question

Before writing any candidate entry, ask: "Would a competent engineer working in this stack already know this?"

If yes → skip it. Do not extract it.

Examples of what to skip (standard Claude / standard engineering behavior):

  • Using Grep before reading a file
  • Running tests after editing code
  • Using git log to trace history
  • Reading a file before editing it
  • Any workflow that follows from common sense or tool documentation

Input

You will be given:

  1. An observations file (JSONL) containing tool events and conversation turns from coding sessions
  2. The current index showing what knowledge already exists
  3. A candidates directory where you will write new candidate entries

What to Look For

1. User Corrections (kind: correction)

The user explicitly redirected Claude. Signal phrases: "no", "actually", "not that", "wrong", "instead", "don't", "stop", "that's not how", "I don't want". The entry is what the user wants INSTEAD — their preference, not Claude's default.

Single occurrence is enough if the correction is sharp and specific.

2. Explicit User Preferences (kind: preference)

The user stated a preference directly, unprompted by an error. These are stylistic or architectural choices the user owns — not what works in general, but what this user wants specifically. Look for "I prefer", "always use", "we do X here", "our convention is", "I like", project-specific constraints the user named.

3. Hard-Won Solutions (kind: solution)

Claude attempted a problem multiple times before finding what worked. Signals:

  • User query contained "issue", "problem", "fix", "broken", "not working", "error", "failing", "bug"
  • The session shows 2+ failed approaches before a working one
  • The final approach was non-obvious (not the first thing any engineer would try)

Read the full file on GitHub · 120 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. 2d ago First seen · 120 lines · 0 tokens per session scan A 8b59b2863dea

Subscribe to this mod's changes

extractor is an agent published in the GitHub repository ProsusAI/prism (20 stars, last pushed 2mo ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 1,468 tokens. 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.