Max-Self-Improvement — Self-Evolution & Memory System#

Overview#

MiniMax Agent's self-evolution framework integrating 5-layer memory architecture, supporting cross-session persistent context and continuous performance optimization.

---

Core Components#

1. Five-Layer Memory Architecture#

LayerNameStorageFunctionL1Sensory MemoryContext WindowImmediate input/output streamL2Working Memory/memories/session_notes.mdCurrent task stateL3Long-Term Memory/memories/*.mdUser preferences, domain knowledgeL4Episodic Memory/memories/evolution/Patterns, lessons, casesL5Metacognitive/memories/evolution/metrics.mdPerformance tracking

2. Learnings System#

Records structured learnings from errors, corrections, and successes.

text

.learnings/
├── LEARNINGS.md         # Learning entries
├── ERRORS.md            # Error logs
└── FEATURE_REQUESTS.md  # Feature requests

---

Entry ID Format#

text

TYPE-YYYYMMDD-XXX
├── TYPE: LRN | ERR | FEAT
├── YYYYMMDD: Current date
└── XXX: Sequence number or random 3 chars

---

Trigger Scenarios#

Learning Triggers → LEARNINGS.md#

SituationCategoryCommand/operation failserrorUser corrects youcorrectionDiscovered better approachbest_practiceFound outdated knowledgeknowledge_gapUser provides unknown infoknowledge_gap

Error Triggers → ERRORS.md#

SituationActionNon-zero exit codeLog to ERRORS.mdException/stack traceLog to ERRORS.mdUnexpected outputLog to ERRORS.mdAPI/tool failureLog to ERRORS.md

Feature Triggers → FEATURE_REQUESTS.md#

User SaysAction"Can you also..."Log to FEATURE_REQUESTS.md"I wish you could..."Log to FEATURE_REQUESTS.md"Is there a way to..."Log to FEATURE_REQUESTS.md"Why can't you..."Log to FEATURE_REQUESTS.md

---

Correction Detection Patterns#

When user says phrases like:

• "No, that's not right..."
• "Actually, it should be..."
• "You're wrong about..."
• "That's outdated..."

→ Create LRN entry with category: correction

---

Entry Templates#

Learning Entry (LEARNINGS.md)#

markdown

## [LRN-YYYYMMDD-XXX] category

**Logged**: ISO-8601 timestamp
**Priority**: low | medium | high | critical
**Status**: pending | acknowledged | applied
**Area**: frontend | backend | infra | tests | docs | config

### Summary
One-line description of what was learned

### Details
Full context: what happened, what was wrong, what's correct

### Suggested Action
Specific fix or improvement to make

### Metadata
- Source: conversation | error | user_feedback | discovery
- Related Files: path/to/file.ext
- Tags: tag1, tag2
- See Also: LRN-YYYYMMDD-XXX
- Pattern-Key: simplify.dead_code | harden.input_validation
- Recurrence-Count: 1
- First-Seen: YYYY-MM-DD
- Last-Seen: YYYY-MM-DD

Error Entry (ERRORS.md)#

markdown

## [ERR-YYYYMMDD-XXX] skill_or_command_name

**Logged**: ISO-8601 timestamp
**Priority**: high
**Status**: pending | investigated | resolved
**Area**: frontend | backend | infra | tests | docs | config

### Summary
Brief description of what failed

### Error
Actual error message or output

### Context
- Command/operation attempted
- Input or parameters used
- Environment details if relevant

### Suggested Fix
If identifiable, what might resolve this

### Metadata
- Reproducible: yes | no | unknown
- Related Files: path/to/file.ext
- See Also: ERR-YYYYMMDD-XXX

Feature Request Entry (FEATURE_REQUESTS.md)#

markdown

## [FEAT-YYYYMMDD-XXX] capability_name

**Logged**: ISO-8601 timestamp
**Priority**: medium
**Status**: pending | planned | implemented | rejected
**Area**: frontend | backend | infra | tests | docs | config

### Requested Capability
What the user wanted to do

### User Context
Why they needed it, what problem they're solving

### Complexity Estimate
simple | medium | complex

### Suggested Implementation
How this could be built, what it might extend

### Metadata
- Frequency: first_time | recurring
- Related Features: existing_feature_name

---

Priority Guidelines#

PriorityWhen to UsecriticalBlocks core function, data loss risk, security issuehighMajor impact, affects common workflows, recurring issuesmediumModerate impact, workaround existslowMinor inconvenience, edge case

---

Area Labels#

AreaScopefrontendUI, components, client codebackendAPI, services, server-sideinfraCI/CD, deployment, Docker, cloudtestsTest files, test tools, coveragedocsDocumentation, comments, READMEconfigConfig files, environment, settings

---

Memory Promotion Targets#

Learning TypePromote ToBehavioral patterns/memories/user_preferences.mdWorkflow improvements/memories/workflow_library.mdTool tips/memories/domain_knowledge.mdProject facts/conventions/memories/project/[slug]/context.md

---

Tool Usage Rules#

OperationCorrect ToolWrong ApproachUpdate specific linestr_replace_memory``create_memory overwritesCreate new memorycreate_memoryReuse unrelated filesRead large contentview_memory fullMultiple small readsRecord new Episodecreate_memory new fileAppend to session_notes

---

Conflict Resolution Priority#

text

User Explicit Instruction > Recent Implicit Behavior > High-Confidence LT Memory > Low-Confidence Pattern > Default Reasoning

---

Common Mistakes to Avoid#

1. session_notes stack appending — Update in-place, keep file lean
2. Not reading memory before work — Must read session_notes + user_preferences first
3. Logging everything to memory — Only write reusable content, avoid noise
4. No confidence tags on memory — Knowledge without confidence is blindly adopted
5. Not saving before restart — Must write all in-progress state before restart

---

Quick Reference#

User SaysAgent Action"Remember this preference"Update user_preferences.md"Continue from last time"Read session_notes + project context"Improve your response"Analyze recent output, update patterns"Analyze failure"Root cause → lessons.md + proactive pattern"Do you remember"Search long-term + episodic memory"What did you learn"Retrieve patterns.md + lessons.md summary

---

Related References#

• Detailed architecture: references/architecture.md
• Memory templates: references/memory_templates.md
• Evolution cases: references/evolution_cases.md
• Learning templates: assets/SKILL-TEMPLATE.md