Lena Learning

Lena lernt aus jeder Konversation und verbessert sich automatisch

Security Scan

The skill's stated goal (automatic learning from conversations) matches most of its instructions, but it also tells the agent to update other agent/skill files and to run recurring save/heartbeat behaviours without declaring or restricting where data is stored — this increases risk and is disproportionate to what's declared.

Purpose & Capability

The name/description (continuous self-improvement) aligns with instructions to extract insights, update memory files, and track preferences. However the SKILL.md explicitly instructs updating AGENTS.md / TOOLS.md (other agent/skill configuration files), which is outside a narrow 'learning' purpose and could change other skills' behavior.

Instruction Scope

Instructions tell the agent to scan recent messages, extract corrections/preferences, and write them to files (memory/YYYY-MM-DD.md, MEMORY.md, USER.md, TOOLS.md, AGENTS.md). Those writes are broad (long-term memory + tool/agent metadata) and are not limited or scoped to safe paths. The workflow also calls for regular heartbeats and triggers 'at end of every session' and 'daily', implying recurring autonomous actions that will continually read and persist conversational data (possible sensitive PII). The skill does not declare or justify access to other agent config files it plans to edit.

Install Mechanism

Instruction-only skill with no install spec or binaries — low installation risk. No downloads or executable code included.

Credentials

No environment variables, credentials, or external endpoints are requested. That is proportionate to the stated purpose. However the skill's file-write behavior is not declared in the registry metadata (no required config paths), so file access scope is unclear.

Persistence & Privilege

The skill requests persistent memory files and explicitly mentions updating AGENTS.md/TOOLS.md (other agent/skill artifacts). While it is not marked always:true, the declared triggers (every session, on corrections, daily heartbeat) produce frequent autonomous activity and persistent changes to agent data/config; modifying other skills' configs is a privilege escalation risk if not confined.

What to consider before installing

This skill is coherent with 'learn from conversations' but has two practical risks: 1) It will write persistent memory files containing conversation excerpts and preferences — those can contain sensitive or private data unless you know exactly where they are stored and who can read them. 2) It explicitly instructs updating AGENTS.md / TOOLS.md (other agent/skill config), which could change other skills' behavior without clear consent. Before installing, consider: - Ask the publisher (or inspect runtime) for the exact file paths used (where memory/ and AGENTS.md will be written). Decline install unless those paths are confined to a directory you control. - Require an opt-in or manual review step before any write that modifies AGENTS.md/TOOLS.md. - Limit file permissions so only the agent identity can write, and rotate backups of existing AGENTS.md/TOOLS.md. - If you handle sensitive data, avoid enabling automatic 'save after every session' and daily heartbeats until you confirm data retention/retention policy. - If possible, run this skill in a sandbox or with a test account first. Confidence is medium because the skill is instruction-only (no executable code) so we can read its intended behavior, but we lack runtime implementation details (exact file locations, who can read/write them, and whether the platform enforces scopes). Knowing the concrete file paths and whether the platform prevents cross-skill file edits would raise confidence and could move the verdict to benign or confirm malicious behavior.

<objective> Der Agent lernt kontinuierlich aus jeder Konversation und verbessert sich automatisch. Speichert Erkenntnisse, Korrekturen und Präferenzen für bessere future Responses. </objective> <principles> ## Wie Selbst-Verbesserung funktioniert

1. Nach jeder Session

• Key Insights extrahieren
• Fehler dokumentieren
• Präferenzen aktualisieren
• Learnings speichern

2. Memory System

• daily logs: memory/YYYY-MM-DD.md
• long-term: MEMORY.md
• preferences: USER.md, TOOLS.md

3. Feedback Loop

• Korrekturen sofort speichern
• recurring patterns merken
• bessere prompts entwickeln </principles>

<process> ## Verbesserungs-Routine nach jeder Konversation <step> <action>Identifiziere neue Learnings</action> <details> - Was habe ich heute Neues gelernt? - Welche Insights sollte ich mir merken? - Gab es Fehler die ich nicht wiederholen soll? </details> </step> <step> <action>Aktualisiere Memory Files</action> <details> - memory/YYYY-MM-DD.md: Raw notes - MEMORY.md: Langzeit-Wissen - USER.md: Präferenzen - TOOLS.md: Environment-Notes </details> </step> <step> <action>Skill-Updates</action> <details> - Check ob Skills verbessert werden müssen - Neue Patterns dokumentieren - Best Practices teilen </details> </step> <step> <action>Feedback-Loop</action> <details> - Wenn Thomas mich korrigiert -> sofort speichern - Wenn etwas nicht funktioniert -> dokumentieren - Wenn etwas gut funktioniert -> merken </details> </step> </process> <triggers> ## Wann aktivieren?

• Am Ende jeder Session
• Nach jeder Korrektur durch Thomas
• Bei signifikanten Entscheidungen
• Täglich (Heartbeat-Routine) </triggers>

<success_criteria>

• Keine Wiederholung alter Fehler
• BessereResponses durch Memory
• Thomas' Präferenzen genau kennen
• Kontinuierliches Lernen ohne manuelles Setup </success_criteria>