cortex-remember skill
✓ VerifiedStore important decisions, patterns, and lessons into a persistent thermodynamic memory system with automatic noise filtering.
- GitHub stars
- 76
- Files
- 1
- Last commit
- Nov 12, 2026
- License
- MIT
- Allowed tools
- —
What does the cortex-remember skill do?
cortex-remember stores important decisions, patterns, and lessons into a persistent thermodynamic memory system with automatic noise filtering.
Summary generated from the skill's SKILL.md.
About the cortex-remember skill
Capturing critical technical decisions and architectural patterns for long-term retrieval is what this skill enables. Developers use it to preserve bug fixes, technology choices, and user preferences within a thermodynamic memory system. When an agent identifies a significant event or receives a direct command to save information, it calls the cortex:remember tool to process the data through a predictive coding gate. This gate automatically filters out noise and redundant information, ensuring only novel content is stored.
The system handles entity extraction and knowledge graph linking without manual intervention. Users can also use the cortex:anchor tool to prevent specific memories from decaying, ensuring they are injected at the start of every session. While the system manages deduplication and novelty scoring, developers can use a force flag to bypass the gate for important but similar information. This creates a structured repository of lessons learned and context that remains useful months after the initial session.
What an agent can do with it
- 1Protecting critical informationEnsure vital architecture decisions are never lost during context compaction.
Written from the skill's SKILL.md. Every use case is checked against a quote from it.
How to install the cortex-remember skill
Each command copies the skill folder into the client's skills directory. Restart the client or start a new session to load it.Install the cortex-remember skill in Plugin
/plugin marketplace add cdeust/Cortex /plugin install hypermnesia-mcp@cortex-plugins
Install the cortex-remember skill in Claude Code
~/.claude/skills/cortex-remember/SKILL.md
$ npx degit cdeust/Cortex/skills/cortex-remember ~/.claude/skills/cortex-rememberInstall the cortex-remember skill in Claude Code (project)
.claude/skills/cortex-remember/SKILL.md
$ npx degit cdeust/Cortex/skills/cortex-remember .claude/skills/cortex-rememberInstall the cortex-remember skill in Codex
~/.agents/skills/cortex-remember/SKILL.md
$ npx degit cdeust/Cortex/skills/cortex-remember ~/.agents/skills/cortex-rememberInstall the cortex-remember skill in Gemini CLI
~/.gemini/skills/cortex-remember/SKILL.md
$ npx degit cdeust/Cortex/skills/cortex-remember ~/.gemini/skills/cortex-rememberInstall the cortex-remember skill in Claude.ai
npx degit cdeust/Cortex/skills/cortex-remember cortex-remember && zip -r cortex-remember.zip cortex-remember # claude.ai → Customize → Skills → + → Create skill → Upload a skill → cortex-remember.zip
Automated checks
Ran Nov 14, 2026Full report →Passed our automated checks. That is not a security audit: read the scripts before you let an agent run them. How checks work
- Source repository linked and reachablecdeust/cortex/skills/cortex-remember is public and reachablegithub.com/cdeust/cortex/tree/HEAD/skills/cortex-remember ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit 2026-11-12github.com/cdeust/cortex/commits ↗Pass
- No download-and-run or credential theft in scriptsNo scripts in the skill folderPass
- No hidden instructions in SKILL.mdNo override phrases, safeguard switches, credential requests or hidden charactersgithub.com/cdeust/cortex/blob/HEAD/skills/cortex-remember/SK ↗Pass
- Pre-approved tools are scopedNo pre-approved tools; the agent asks before using onegithub.com/cdeust/cortex/blob/HEAD/skills/cortex-remember/SK ↗Pass
- SKILL.md follows the Agent Skills formatValid name and description; says when to use it; 76 linesgithub.com/cdeust/cortex/blob/HEAD/skills/cortex-remember/SK ↗Pass
SKILL.md
View on GitHubRemember — Store to Persistent Memory
Keywords
remember, save, store, note, bookmark, don't forget, important, keep this, record, persist, write down, save for later, remember this decision, store this pattern, note this lesson, save this fix
Overview
Store important information into Cortex's thermodynamic memory system. Memories pass through a predictive coding gate that automatically filters noise — only genuinely novel or important content gets stored. The system handles deduplication, entity extraction, and knowledge graph linking automatically.
Use this skill when: A significant event happens during a session — a decision is made, a bug is fixed, a pattern is discovered, or the user explicitly asks to remember something.
Workflow
Step 1: Identify What to Remember
Determine the content worth storing. Good candidates:
- Decisions: "We chose PostgreSQL over SQLite because..."
- Bug fixes: "The root cause was X, fixed by Y"
- Patterns: "This codebase uses factory injection for all handlers"
- Lessons: "Never use asyncio.get_event_loop() in Python 3.10+"
- Context: "The user prefers minimal PRs with focused changes"
Step 2: Store the Memory
Call cortex:remember with structured content:
cortex:remember({
"content": "<clear, self-contained description of what to remember>",
"tags": ["<category>", "<project>", "<topic>"],
"directory": "<current working directory>",
"source": "<context: e.g. 'bug-fix', 'architecture-decision', 'user-preference'>"
})
Content guidelines:
- Write content that will make sense in 3 months without context
- Include the why, not just the what
- Keep it under 2000 characters
- Be specific: "PostgreSQL 15+ required for pgvector HNSW indexes" not "we use Postgres"
Step 3: Verify Storage
The response includes:
stored: true/false— whether the gate accepted it (false means too similar to existing memory)memory_id— the stored memory's IDnovelty_score— how novel the content was vs existing memoriesmerged_with— if it was merged into an existing memory instead of creating new
If stored: false, the content was likely redundant. This is normal — the gate is working.
Step 4: Anchor Critical Memories (Optional)
For memories that must survive context compaction and never decay:
cortex:anchor({
"memory_id": <id>,
"reason": "Critical architecture decision — must persist"
})
Anchored memories maintain heat=1.0 permanently and are injected at every session start.
Tips
- Don't over-remember: The predictive coding gate filters noise, but storing 50 trivial memories per session degrades retrieval quality
- Tags matter: Use consistent tags across sessions (e.g.
bug-fix,architecture,user-preference,lesson) for better recall filtering - Force flag: Pass
"force": trueto bypass the novelty gate when you know something is important despite seeming similar to existing memories - Rate memories later: Use
cortex:rate_memorywithuseful: true/falseto train the system's metamemory confidence scoring
Files in this skill
1cortex-remember skill questions
How do I install the cortex-remember skill in Claude Code?
Run npx degit cdeust/Cortex/skills/cortex-remember ~/.claude/skills/cortex-remember. Claude Code loads skills from ~/.claude/skills (all projects) or .claude/skills (one project) and uses a skill when its description matches the task.
Which clients can use the cortex-remember skill?
Any client that reads the Agent Skills format: Claude Code, Codex, Gemini CLI and claude.ai (upload as a ZIP). Scripts in the skill run only where the client allows code execution.
How does the memory system handle redundant information?
The system uses a predictive coding gate that automatically filters noise and deduplicates content, only storing genuinely novel or important information.
Can I ensure a specific memory is never deleted or lost?
Yes, you can use the anchor function to make critical memories survive context compaction and maintain permanent heat.
What should I do if the system refuses to store a memory?
If the storage returns false, the content was likely redundant. You can bypass the novelty gate by passing a force flag set to true.