// MUST use when user says 'continuous-improvement', 'instinct status', 'what have you learned', 'show learned rules', 'mulahazah status', 'what patterns have you noticed', 'behavioral learning', 'show rules', or when user asks about session learning, accumulated rules, or patterns Claude has observed.
MUST use when user says 'save topic', 'save to topic diary', 'remember this under', 'review topic', 'list topics', or when a generic 'save' request needs routing between session memory, daily diary, topic diary, or all targets.
Auto-triggers when user says 'new adventure', 'start adventure', 'new fantasy', 'save adventure', 'end adventure', 'load adventure', 'resume adventure', 'continue adventure', 'let's play', 'VN mode', or on numbered choice input (1-5) or free text dialogue during an active adventure.
Auto-triggers when user says 'create songs', 'new album', 'create album', 'make music', 'muse this', 'write a song', 'create a song', 'compose', 'song from image', 'album from image', 'generate album', 'write songs', or when user shares an image and wants to create music from it.
Auto-triggers when user asks for a Midjourney or NijiJourney image prompt, when creating visual art prompts, or when user says 'midjourney prompt', 'niji prompt', 'create prompt', 'create a prompt', 'image prompt', 'generate prompt', 'draw this', 'reference sheet', 'make an image of'. Generates optimized AI image prompts with composition-aware framing.
MUST use when user says 'survey project', 'scan project', 'check health', 'investigate', 'deep dive', 'what's going on in', 'look into', 'refine code', 'clean up code', 'review changes', 'sharpen', 'audit system', 'full audit', 'show me everything', or when the AI needs to assess project health before planning, review code quality after implementation, or investigate a bug. Also triggers on 'how does this project look', 'what's the status', 'review what I changed', 'check for issues'.
Auto-triggers when AI detects a repeated pattern handled ad-hoc 3+ times, when AI makes a mistake that a permanent rule would prevent, when AI identifies a workflow that should be automated as a skill, or when user says 'create skill', 'new skill', 'forge this', 'level up', 'upgrade skill', 'self improve', 'improve skill'. Also triggers when AI wants to propose a level-up to an existing skill based on conversation patterns.
| name | mulahazah |
| description | MUST use when user says 'continuous-improvement', 'instinct status', 'what have you learned', 'show learned rules', 'mulahazah status', 'what patterns have you noticed', 'behavioral learning', 'show rules', or when user asks about session learning, accumulated rules, or patterns Claude has observed. |
The AI that remembers how you work.
When this skill activates, output:
Mulahazah active -- reading learned rules for this project...
Then execute the protocol below.
Mulahazah is a passive observation layer built into your AI companion. It captures every tool call as structured data, analyzes patterns with Haiku, and writes actionable rules to rules.md. Those rules persist across sessions -- silently shaping how your AI works with you, without needing to be reminded each time.
| Context | Status |
|---|---|
| User asks about learned rules or patterns | ACTIVE -- full protocol |
User types /continuous-improvement | ACTIVE -- full protocol |
| User says "what have you learned" | ACTIVE -- full protocol |
| User wants to see rule summary | ACTIVE -- full protocol |
| General coding conversation | DORMANT -- follow rules.md silently |
| User is explicitly asking about Forge skills | DORMANT -- use Forge skill instead |
~/.claude/mulahazah/rules.md -- this is where learned rules live/continuous-improvement after a work session to generate them."Generate a reflection for this session:
## Reflection -- [Date]
- What worked:
- What failed:
- What I'd do differently:
- Rule to add:
If there is a "Rule to add", check ~/.claude/mulahazah/rules.md to see if it already exists. If not, append it.
~/.claude/mulahazah/projects/*/observations.jsonlbash ~/.claude/mulahazah/bin/analyze.sh~/.claude/mulahazah/observer.pid exists and process is alivebash ~/.claude/mulahazah/agents/start-observer.shRead ~/.claude/mulahazah/rules.md and display:
~/.claude/mulahazah/rules.md.~/.claude/mulahazah/ does not exist, output the installation prompt and stop.| Situation | Behavior |
|---|---|
| No rules.md yet | Report "No rules yet for this project. Run /continuous-improvement after your first work session." |
| Mulahazah not installed | Output: "Mulahazah is not installed. See Feature/Mulahazah-System/install-mulahazah.md to get started." |
| Observer not running | Note it is optional and show start command. Do not block skill execution. |
| analyze.sh not found | Report the missing path and remind user to re-run install steps. |
| Fewer than 5 observations | Note more tool calls are needed before patterns can be extracted. |
| Feature | How Mulahazah Interacts |
|---|---|
| Forge | Rule clusters from rules.md with repeated patterns become Forge skill proposals. |
| Save Diary | Active rules from rules.md are summarized in the session diary. |
| Memory Consolidation | Triggers review of rules.md for stale or contradictory entries. |
| Decision Log | Rules applied during a session are logged as behavioral decisions. |
observe.sh, Haiku-powered analysis via analyze.sh, /continuous-improvement command. Background observer optional. (Origin: continuous-improvement v2.0, ported to MemoryCore as Mulahazah System. Tested end-to-end: observations captured, analyze.sh extracted "Grep → Read → Edit workflow" rule.)