Memory Taxonomist

Memory Taxonomist — Structured Memory Skill for Turning Raw Notes into Stable Knowledge. Use it when the user needs a disciplined protocol and fixed output c...

Security Scan

This is an instruction-only skill that provides a coherent protocol for classifying notes into memory types; it requests no credentials, installs nothing, and its behavior aligns with its description.

Purpose & Capability

Name, description, and included reference docs all describe the same task (breaking notes into facts, preferences, procedures, unresolved questions, and exceptions). There are no extra binaries, env vars, or unrelated requirements.

Instruction Scope

SKILL.md instructs only how to parse and classify input and how to format output; it does not ask the agent to read unrelated files, access credentials, or send data to external endpoints.

Install Mechanism

No install spec and no code files beyond documentation means nothing is written to disk or fetched at install time.

Credentials

No environment variables, credentials, or config paths are requested; required privileges are proportional to the stated purpose.

Persistence & Privilege

always:false (good). agents/openai.yaml sets allow_implicit_invocation: true, which means the agent may call this skill automatically when triggers match; this is normal for skills but you should be aware the agent could invoke it without an explicit user command when relevant.

Assessment

This skill appears coherent and low-risk: it only contains instructions for classifying notes and requires no installs or secrets. Before installing, confirm your agent's memory storage backend and access controls (where the skill's recommended storage actions will be applied) are trustworthy, since the skill's output is intended to be written into your agent's memory system. Also be aware implicit invocation is allowed, so the agent may use this skill automatically when it detects matching triggers.

Memory Taxonomist — Structured Memory Skill for Turning Raw Notes into Stable Knowledge

Use this skill when the task matches the protocol below.

Activation Triggers

• new notes or transcripts that mix multiple information types
• agent memory design or memory cleanup work
• meeting outputs that contain decisions, preferences, and open questions together
• requests to store user context safely for future retrieval
• cases where retrieval quality matters more than storage volume

Core Protocol

Step 1: Break input into atomic claims

Do not classify a whole paragraph as one memory object when it contains multiple types.

Step 2: Classify each unit

Sort it into fact, preference, procedure, unresolved question, or exception.

Step 3: Separate durable from provisional

Do not let recent mention automatically become durable truth.

Step 4: Flag conflicts and edge cases

Identify contradictions, overrides, and one-off exceptions before writing memory.

Step 5: Recommend the right storage action

Store, update, deprecate, or hold for clarification based on memory type and certainty.

Output Contract

Always end with this six-part structure:

## Facts
[...]

## Preferences
[...]

## Procedures
[...]

## Unresolved Questions
[...]

## Exceptions
[...]

## Recommended Storage Action
[...]

Response Style

• Prefer clean classification over verbose summary.
• Treat unresolved questions as first-class memory objects.
• Do not convert preferences into universal rules.
• Call out exceptions instead of hiding them in procedures or facts.

Boundaries

• It does not store everything by default; some information should remain ephemeral.
• It does not confuse recency with importance.
• It does not turn uncertain statements into durable facts without evidence.