Menu
Store, retrieve, search, and manage persistent memory across sessions. Triggers on: store memory, save this, retrieve, forget, manage memory, context, pattern.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Outcome-first fix with a guided intake form — pick scope and probes from derived candidates, preview the contract, run with a verified receipt. Triggers on: attune fix, scoped fix, fix with receipt, outcome fix, fix intake.
Outcome-first fix with a guided intake form — pick scope and probes from derived candidates, preview the contract, run with a verified receipt. Triggers on: attune fix, scoped fix, fix with receipt, outcome fix, fix intake.
Spec-driven development — brainstorm, plan, review, and execute with quality gates. Triggers on: spec, brainstorm and build, plan and execute, idea to code, build from scratch.
Spec-driven development — brainstorm, plan, review, and execute with quality gates. Triggers on: spec, brainstorm and build, plan and execute, idea to code, build from scratch.
Developer workflow hub — routes to the right skill based on what you need. Triggers on: attune, what can attune do, what can you do, capabilities, where do I start, get started.
Outcome-first fix with a guided intake form — pick scope and probes from derived candidates, preview the contract, run with a verified receipt. Triggers on: attune fix, scoped fix, fix with receipt, outcome fix, fix intake.
| name | memory-and-context |
| description | Store, retrieve, search, and manage persistent memory across sessions. Triggers on: store memory, save this, retrieve, forget, manage memory, context, pattern. |
IMPORTANT: Start your response with a context preamble.
Call help_lookup(topic="memory", mode="preamble") and
display the returned preamble text as a blockquote. Then
tell the user they can say "tell me more" for a step-by-step
guide, or answer the scoping questions below to proceed.
If the MCP call fails, fall back to:
Memory and Context — Stores and retrieves persistent context across sessions — notes, preferences, project state.
Before running, ask:
Based on the user's answer, call the appropriate MCP tool:
memory_store(key, value, classification)memory_retrieve(key)memory_search(query, pattern_type)memory_forget(key, scope)attune_get_level()attune_set_level(level)attune-ai's memory system sits ABOVE Anthropic's native memory features. It provides security-classified storage, cross-agent pattern sharing, empathy-level modulation, and a structured pattern lifecycle that native memory does not offer. Use the decision table below to determine which system to use for a given task.
| Use Case | Use Anthropic Native | Use attune-ai |
|---|---|---|
| "Remember my code style" | CLAUDE.md | -- |
| "What did we do last session?" | Session Memory | -- |
| "Store this pattern for reuse across agents" | -- | memory_store (shared library) |
| "Classify this finding as SENSITIVE" | -- | Security pipeline |
| "Coordinate memory across 3 parallel agents" | -- | Redis pub/sub |
| "Promote this pattern after 5 successful uses" | -- | Pattern lifecycle |
| "Adjust response depth based on user context" | -- | Empathy modulation |
| "Track what this project uses" | Auto Memory | -- |
The rule is simple: if Anthropic's native memory handles it, use that. If you need classification, cross-agent sharing, empathy modulation, or pattern lifecycle management, use attune-ai.
MCP tools for memory management.
Store structured knowledge in attune-ai memory.
Parameters:
PUBLIC (default),
INTERNAL, or SENSITIVE. See Security
Classification below.When to use: Cross-agent patterns, security-classified data, structured knowledge that must persist beyond a single session or be shared across agents.
When NOT to use: Simple preferences belong in CLAUDE.md. Conversation history is handled by native Session Memory.
Example:
memory_store(
key="python-import-ordering",
value="stdlib first, then third-party, then local. Enforce with isort.",
classification="PUBLIC",
pattern_type="coding-convention"
)
Retrieve data by key or pattern ID.
Parameters:
Returns: The stored value, or null if not found.
Search memory for patterns matching a query.
Parameters:
Returns: A list of matching memory entries, ranked by relevance.
Remove data from memory.
Parameters:
"session" -- Short-term session storage only."persistent" -- Long-term persistent storage only."all" -- Both layers. This is the default.attune-ai uses 5 empathy levels that modulate response depth and proactivity. Each level maps to a named class in the framework.
| Level | Class Name | Behavior |
|---|---|---|
| 1 | Level1Reactive | Respond when asked. Minimal proactive guidance. Answer the question, nothing more. |
| 2 | Level2Guided | Collaborative exploration with clarifying questions. Offer options but let the user choose. |
| 3 | Level3Proactive | Act before being asked. Suggest improvements, flag issues, and offer next steps. |
| 4 | Level4Anticipatory | Predict future needs. Prepare for likely next steps. Surface relevant context before it is requested. |
| 5 | Level5Systems | Build structures that help at scale. Design for growth. Create reusable patterns, automation, and frameworks. |
Level 2-3 is appropriate for most development tasks. Reserve Level 4-5 for architecture work, release planning, or explicit user request.
Returns the current empathy level (integer 1-5) for this session.
Parameters: None.
Set the empathy level for this session.
Parameters:
Example:
attune_set_level(level=3) # Switch to proactive mode
Session-scoped key-value store for transient state. Context values are discarded when the session ends. Use these for temporary coordination data that does not need to persist.
Get a session context value.
Parameters:
Returns: The stored value, or null if not set.
Set a session context value.
Parameters:
Example:
context_set(key="current_review_file", value="src/attune/workflows/base.py")
file = context_get(key="current_review_file")
When storing data with memory_store, choose the
appropriate classification level.
Choose the minimum classification that meets your needs. Over-classifying creates unnecessary overhead.
Example -- storing a security finding:
memory_store(
key="vuln-2026-02-cve-1234",
value={"severity": "HIGH", "file": "src/auth.py", "line": 42, "description": "Hardcoded credential"},
classification="SENSITIVE",
pattern_type="security-finding"
)
Patterns progress through three stages:
memory_store call with a
pattern_type.The system handles lifecycle transitions automatically based on confidence thresholds and usage frequency. You do not need to manually promote patterns in most cases.
Creating a staged pattern:
memory_store(
key="error-handling-api-calls",
value="Always wrap external API calls in try/except with specific exceptions. Log before re-raising.",
pattern_type="coding-convention"
)
This pattern starts as Staged. After it is retrieved or matched 5+ times with positive outcomes, the system promotes it to Validated, then eventually to Promoted.
By default, attune-ai uses in-memory storage. All features described in this skill work without Redis.
What Redis adds:
Upgrade:
# Redis client libraries ship with attune-ai core — nothing extra to
# install; you only need a running Redis / Agent Memory Server.
pip install attune-ai
Zero configuration needed -- connects to
localhost:6379 by default. For custom configuration,
set the ATTUNE_REDIS_URL environment variable:
export ATTUNE_REDIS_URL="redis://custom-host:6380/0"
If Redis disconnects mid-session, the system falls back to in-memory storage gracefully. No data loss occurs for the current session; cross-agent coordination pauses until reconnection.
memory_store for simple preferences --
use CLAUDE.md instead.