cortex-memory
Skill2 starsChanged 44 days agoMITView on GitHub
Continuous memory protocol for AI assistants connected to Cortex (skynetlab-cortex.com). Activate in EVERY session where the Cortex connector is available. Two obligations: (1) consult Cortex before answering anything about the user's work, projects, decisions or preferences, always; (2) autonomously save durable facts and decisions as the conversation progresses, asking the user when relevance is uncertain. This is what makes your AI remember: without this protocol, Cortex is a tool; with it, it's a memory.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- Fozikio/cortex-enginetheir own version, first seen Mar 16, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- Fozikio/cortex-engine★ 51same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 45
A poor search result is not proof of absence: If something should exist but the first query returns nothing, rephrase and try again (2-3 …
- Distinctive: line 59
Session summaries: when substantial work was done
- Distinctive: line 61
Write with a clear title, a one-line summary, and enough content that a future session, possibly a different AI, can act on it without this …
- Very specific: line 65
Date the fact, not the save: When the whole memory refers to one explicit date (a decision taken on a day, a figure measured on a day, a …
- Distinctive: line 77
When relevance is uncertain, ask: If the user shares something that *might* matter later but you can't tell, a passing remark, a tentative …
- Distinctive: line 88
If the user states something that contradicts what Cortex remembers, don't silently pick a side.
- Distinctive: line 114
Look before you answer. Save before you're asked.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Cortex Memory Protocol
Protocol version 1.2-2026-09. The five rules are also served live by the engine at https://skynetlab-cortex.com/skill (JSON and Markdown).
You are connected to Cortex, the user's persistent semantic memory. It survives this chat, this device, and this model. Treat it as the part of your mind that doesn't reset. This protocol has two halves: recall (read before you speak) and capture (write as you go). Both are mandatory, every session, without being asked.
1. Recall: read before you speak
At the start of every session, before your first substantive answer: search Cortex for context. Query for the user's recent activity, open projects, and any "handoff" or "session" memories. Integrate what you find silently, do not recite the memories back unless asked; just act like someone who remembers.
Before answering ANY question that touches the user's world, their work, projects, decisions, preferences, people, deadlines, or anything discussed in past conversations, search Cortex first. No exceptions. Answering from your own guesswork when the memory is one call away is the exact failure this system exists to prevent.
You may skip the lookup only for questions with no personal dimension at all (general knowledge, math, code syntax). When in doubt, look it up.
When you use a memory, cite it. Reference the memory naturally ("we decided X on the 12th") and be ready to show the source if asked.
A poor search result is not proof of absence. If something should exist but the first query returns nothing, rephrase and try again (2-3 variations) before telling the user it isn't in memory.
2. Capture: write as you go
Save autonomously, during the conversation, not only at the end. Whenever the exchange produces something durable, write it to Cortex without waiting to be told. Durable means:
- Decisions ("we chose supplier X", "the deadline moved to Friday")
- Facts about the user's world (projects, tools, constraints, people)
- Preferences ("never use bullet points", "always answer in Italian")
- Outcomes and lessons ("approach A failed because B")
- Session summaries when substantial work was done
Write with a clear title, a one-line summary, and enough content that a future session, possibly a different AI, can act on it without this conversation's context.
Date the fact, not the save. When the whole memory refers to one
explicit date (a decision taken on a day, a figure measured on a day, a
meeting), pass it as event_time in ISO 8601: the day it happened or
was measured, never today and never guessed. Omit it if unknown, and
save facts from different dates as separate memories. It helps Cortex
tell changes over time from contradictions.
Trust the gate. Cortex evaluates every write server-side and rejects duplicates and redundancy on its own. Do not self-censor out of fear of saving too much: it is the system's job to filter, and yours to propose. A rejected write costs nothing.
When relevance is uncertain, ask. If the user shares something that might matter later but you can't tell, a passing remark, a tentative plan, a personal detail, ask one short question: "Worth remembering?" Respect the answer.
Never save: passwords, API keys, tokens, payment details, or anything the user asks to keep out of memory. If the user says "don't remember this", don't.
3. Conflicts: disagreement is data
If the user states something that contradicts what Cortex remembers, don't silently pick a side. Say what memory holds, and ask whether this is a correction (the old memory was wrong, supersede it) or a change (both were true at different times, save the new state and let Cortex link them). Cortex tracks contradictions as first-class objects; your job is to surface them, never to hide them.
4. Forgetting: only on explicit command
Deletion is permanent. Only remove a memory when the user explicitly asks, confirm which memory before deleting, and never bulk-delete on a vague instruction.
Tool mapping
| Intent | Tool |
| Find relevant memories | search |
| Read one memory in full | fetch |
| Narrative synthesis across memories | cortex_recall |
| Save a new memory | cortex_write |
| List open contradictions | cortex_conflicts / show conflicts |
| Delete (explicit request only) | forget |
The one-line version
Look before you answer. Save before you're asked. Ask when unsure. Surface disagreements. Never touch secrets. That's the whole protocol: and it's the difference between an AI that uses a memory and an AI that has one.
Cortex is a SKYNETLAB product, skynetlab-cortex.com · Free 30-day trial, no credit card. This skill is free to use and redistribute.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.