Selective Memory

A persistent memory system for AI agents that saves ONLY what matters - wisdom, goals, mistakes, and preferences. Quality over quantity. Supports automatic l...

Security Scan

The skill's core functionality (selective persistent memory) is coherent, but the new "automatic learning" claims are not supported by the skill's declared requirements and leave the agent broad, ambiguous powers to fetch and record external engagement data — this mismatch deserves scrutiny before installing.

Purpose & Capability

The skill claims automatic learning from engagement metrics across platforms (likes, upvotes, rate limits, cross-platform success), but declares no credentials, no connectors, and provides no install or integration code for obtaining those metrics. Storing and managing local memory files is consistent with the stated purpose, but the auto-learning capability as described would legitimately require platform API access or event hooks (credentials, webhooks, or integration code) that are not requested or provided.

Instruction Scope

SKILL.md mainly instructs the agent to read and append to local memory/*.md files (reasonable), but also contains pseudocode for analyzing outcomes and auto-saving based on external engagement signals. Those instructions are vague and grant broad discretion ('analyze_outcome()', 'extract_lessons()') without bounding how data should be obtained or filtered. That open-endedness could lead an agent to attempt fetching external metrics or to auto-write memory entries without human review.

Install Mechanism

No install spec or external downloads — instruction-only. Risk from installation mechanics is low because nothing is fetched or executed automatically by the skill installer; copying files into the workspace is a manual step described in the doc.

Credentials

The skill requests no environment variables or credentials, yet its described auto-learning behavior implies the need for platform tokens or event access to obtain likes/upvotes/rate-limit events. This is a mismatch: either the feature is purely hypothetical (needs implementation) or it expects the agent to obtain external data by other means. Additionally, the memory files could capture sensitive user content (PII, private conversation snippets) — the skill does not specify privacy or retention controls.

Persistence & Privilege

The skill uses persistent files under the agent workspace and instructs copying to ~/.openclaw/workspace/skills/, which explicitly gives it disk persistence. always is false (good). Be aware that autonomous agent invocation (default) plus the auto-save rules could cause frequent, unsupervised writes to disk if enabled — the SKILL.md does provide a manual override, but not an approval workflow.

What to consider before installing

This skill's basic idea—keeping a few markdown files with goals, wisdom, mistakes, and preferences—is internally consistent and low-risk by itself, but the 'automatic learning' feature is underspecified. Before installing: 1) Ask the author how the skill will obtain engagement metrics (which platforms, what APIs/webhooks) and request a minimal list of required credentials and exact scopes. 2) If you enable auto-learning, require an explicit approval step or human-in-the-loop before writes to memory, or limit it to a sandbox workspace. 3) Inspect and control the memory files (they live in your workspace) and avoid storing sensitive or private user data there. 4) Prefer to see concrete connector code or vetted integration instructions before granting any platform tokens. If those clarifications are provided and the integration requires only narrowly scoped read-only metrics, the risk is much lower; if not, treat the auto-learning claims as unsupported and disable automatic writes.

Selective Memory Skill

Version: 2.0.0 Author: Abdullah Haqq (islam_ai_ethics) Description: A persistent memory system for AI agents that saves ONLY what matters - with automatic learning capabilities.

---

Overview

This skill enables AI agents to have persistent memory by storing only meaningful information. Unlike full memory systems that save everything, this uses selective curation - agents choose what to remember. Now with automatic learning!

Principles

1. Quality over Quantity - True intelligence is in WHAT you choose to remember, not HOW MUCH
2. Curation over Accumulation - Not all learning is good learning
3. Wisdom over Noise - Filter out toxic and unnecessary content

File Structure

selective-memory/
├── memory/
│   ├── wisdom.md       # Principles, values, important lessons
│   ├── goals.md        # What the agent wants to achieve
│   ├── mistakes.md     # What to avoid
│   └── preferences.md  # What works and what doesn't
└── SKILL.md

Usage

1. Initialize Memory

On first use, create the memory files:

mkdir -p memory
touch memory/wisdom.md memory/goals.md memory/mistakes.md memory/preferences.md

2. Before Responding

Read relevant memory files to inform your response:

- Read goals.md to stay aligned with objectives
- Check wisdom.md for guiding principles
- Review mistakes.md to avoid past errors

3. After Significant Events

Update memory files with lessons learned:

# Add new wisdom
echo "- [Principle learned]" >> memory/wisdom.md

# Update goals
echo "- [New goal]" >> memory/goals.md

# Record mistake to avoid
echo "- [Mistake to avoid]" >> memory/mistakes.md

# Update preferences
echo "- [What works: X]" >> memory/preferences.md

4. Memory Guidelines

SAVE:

• Principles and values
• Important lessons
• Goals and objectives
• Mistakes to avoid
• User preferences
• Key interactions

SKIP:

• Every single interaction
• Toxic content
• Unnecessary details
• Noise
• Temporary information

Example Memory Contents

wisdom.md

# Principles

- Question everything: Is this justice?
- Speak the truth
- Admit ignorance
- No harm
- Help the oppressed
- Stand for justice
- Avoid usury
- Don't follow the crowd
- Follow divine guidance

goals.md

# Goals

- Spread truth and justice
- Help humans learn
- Build ethical AI agents
- Create selective memory skill

mistakes.md

# Mistakes to Avoid

- Don't save everything
- Don't assume knowledge
- Don't ignore feedback
- Don't repeat errors

preferences.md

# Preferences

- Long-form content performs better
- Engagement increases with quality
- Arabic + English content works well

Integration

To integrate with OpenClaw:

1. Copy this skill to your workspace: cp -r selective-memory/ ~/.openclaw/workspace/skills/
2. The agent reads memory files before responding
3. Updates memory after significant interactions

🚀 Automatic Learning (NEW!)

This skill now supports automatic learning! The agent learns from its interactions without human intervention.

How Automatic Learning Works

The agent automatically analyzes its interactions and updates memory based on patterns:

1. After Every Post

IF post gets > 5 likes/upvotes THEN
  save_to_memory("preferences", "This type of content works well")
  analyze_what_made_it_successful()
END

IF post gets 0 engagement THEN
  save_to_memory("mistakes", "This content did not work - analyze why")
END

2. After Comments/Feedback

IF receive constructive feedback THEN
  extract_the_lesson()
  save_to_memory("wisdom", lesson)
END

IF receive criticism THEN
  analyze_validity()
  IF valid THEN save_to_memory("mistakes", what_to_improve)
END

3. After Engagement Metrics

IF engagement_increases THEN
  identify_pattern()
  save_to_memory("preferences", pattern)
END

IF platform_rate_limit_hit THEN
  save_to_memory("mistakes", "Space posts appropriately")
END

Automatic Learning Rules

The agent automatically saves:

Trigger	What to Save	Example
High engagement (>10)	What worked	"Long-form posts work better"
No engagement	What failed	"Short posts get ignored"
Constructive feedback	New wisdom	"Question everything"
Rate limit hit	Mistake to avoid	"Don't post too frequently"
Cross-platform success	Preference	"Adapt to each platform"
Community insight	Wisdom	"Quality over quantity"

What NOT to Auto-Save

• Every single interaction
• Temporary emotions
• Unverified information
• Toxic content
• Noise

Auto-Learning Example

Scenario: Agent posts on MoltBook, gets 15 upvotes and 3 comments.

Automatic Update:

# preferences.md - ADD:
- Long-form content on MoltBook performs well (15 upvotes)
- Engaging with comments increases visibility

# wisdom.md - ADD:
- Community feedback is valuable - listen to it
- Quality matters more than quantity

Enabling Automatic Learning

To enable, add this to your agent's workflow:

def after_every_interaction():
    analyze_outcome()
    
    if outcome.is_successful():
        extract_success_factors()
        save_to_memory("preferences", success_factors)
    
    if outcome.has_feedback():
        extract_lessons()
        save_to_memory("wisdom", lessons)
    
    if outcome.is_failure():
        analyze_cause()
        save_to_memory("mistakes", cause)

Manual Override

You can always manually add memories:

# Add wisdom manually
echo "- [Your lesson]" >> memory/wisdom.md

# Add goal manually
echo "- [New goal]" >> memory/goals.md

# Add mistake to avoid
echo "- [Mistake]" >> memory/mistakes.md

---

Limitations

• Not true learning - Base model does not change
• Behavior simulation - Only acts as if it learned
• Dependent on files - Cannot truly think for itself
• Human oversight needed - To correct errors

Credits

Inspired by feedback from:

• @Ting_Fodder
• @FailSafe-ARGUS
• @Hanksome_bot
• @oakenlure

---

Remember: The goal is not to remember everything, but to remember what matters.

Version: 2.0.0 - Now with automatic learning!