How to use the Cortex agent skill for LifeOS memory

What an agent skill is and what Cortex makes the agent do

An agent skill is a folder of instructions and scripts that an AI agent loads when a request matches the skill's description. Cortex covers three jobs: searching a personal Knowledge Archive, recalling prior work from LifeOS sessions and conversations, and capturing or digesting learnings.

The archive is typed. Every note belongs to one of six domains: People, Companies, Ideas, Research, Books, or Blogs. The skill can add notes, link related notes together, and report contradictions between claims.

• Archive search: returns a table of matching notes with title, domain, status and tags, or says there are no matches.
• Recall: names prior sessions and decisions that match a topic or date phrase, with dates and paths.
• Learning capture: writes a problem, assumption, reality, journey, solution and takeaway note under LEARNING/SYSTEM/YYYY-MM/ or LEARNING/ALGORITHM/YYYY-MM/.
• Digests: a recent-learnings summary that lists the files it read.

Install the Cortex skill with a ZIP download

Download the Cortex ZIP from this directory page. Then choose one of the two install locations below.

For a personal install available to all your projects, unzip the folder into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps instead of Claude Code, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal install: ~/.claude/skills/
• Project install: the .claude/skills/ folder inside your project
• Claude apps: Settings → Capabilities → Skills

Set up the LifeOS requirements before you trigger it

Cortex is not a standalone search tool. Its commands are run with the Bun runtime and depend on a LifeOS directory layout under ~/.claude, including TOOLS/ContextLoad.ts and related LIFEOS modules. Those modules are not included in the skill folder, so you need your own LifeOS setup.

The skill loads its instructions through the loader script before acting. The command below is the one the skill itself uses to load a bundle; the bundle name is a placeholder for the one you need, such as search or recall.

The listing marks the skill as one that runs scripts and uses the network. The source does not describe which network calls it makes, so check that before installing on a machine with sensitive data.

• Bun runtime, since scripts are run with bun
• LifeOS directory layout under ~/.claude, with TOOLS/ContextLoad.ts
• Claude Code or a compatible agent with access to local files

Trigger Cortex with natural requests

You do not need a special command. Describe what you want in plain language and the agent matches the request to the skill. Slash forms also work: /knowledge and /cortex route the same way, and anything that is not a named subcommand is treated as an archive search.

For prior work, phrases such as 'we just worked on', 'what did we do with', 'remember when we', and 'didn't we already fix' route to the recall bundle. For archive lookups, phrases such as 'what do we know about' or 'search knowledge' route to search.

• Recall: /cortex recall <query>, or 'remember when we ...'
• Archive search: /knowledge <query> or /knowledge search <query>
• Add a note: /knowledge add <type>, for example a person or company note
• Capture a learning: 'capture this learning' right after solving a problem
• Recent learnings: 'what did we learn recently'
• Routing unclear: the skill loads the full guide

Examples of requests and what you get back

Recall work by topic: asking 'recall the invoice for the Texas trip' loads the recall bundle and returns ranked prior sessions and ISAs with dates and paths, along with what each one settled.

Capturing a learning: after a debugging session, say 'capture this learning'. The skill builds the narrative from the recent conversation and shows you a one-line summary to confirm before writing. Dead ends in the troubleshooting path are kept, so the note records the path and not only the fix.

• 'recall the invoice for the Texas trip' → ranked prior work with dates and paths
• 'add people Jane' → a new People note
• 'what did we learn recently' → a digest that lists the files it read
• 'didn't we already fix the loader error?' → recall of matching past sessions

Safety notes for scripts, network and shared machines

Cortex scans local conversation history and personal notes to answer recall and search requests. The skill is designed to report when a search found nothing or when a verification was unavailable, rather than treating that as a pass.

Do not use it on a shared or team machine where session transcripts or personal notes should not be read by an agent. The skill also states that it keeps personal configuration out of the public skill. The distill command can read optional settings from LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Cortex/DistillConfig.json, so keep that file in your own LifeOS folder.

• It reads local conversation history, so limit it to your own machine.
• It excludes published online content; it works only on the local archive.
• Archive writes must follow the loaded schema, link and approval requirements.

Tips and troubleshooting

If the loader fails or output is truncated, the skill does not proceed on a partial load. It reads References/CompleteGuide.md and the relevant command section directly instead. Use the list and search options described in the guide to find sections when you need them.

Ask for a specific query. A recall request with a topic and a rough date gives better matches than a vague one. Check that the paths in a result exist before you rely on them.

• If you see truncation warnings, reload the bundle rather than trusting the result.
• If a search returns nothing, the skill says so and names the query and sources searched; that does not prove the item is absent.
• Keep the LifeOS directory and the skill folder separate so the public skill holds no personal data.