Munger Observer

Daily wisdom review applying Charlie Munger's mental models to your work and thinking. Use when asked to review decisions, analyze thinking patterns, detect biases, apply mental models, do a "Munger review", or run the Munger Observer. Triggers on scheduled daily reviews or manual requests like "run munger observer", "review my thinking", "check for blind spots", or "apply mental models".

Security Scan

The skill's requests and instructions are consistent with its stated purpose (reviewing a user's recent activity and applying mental models), but it explicitly instructs the agent to read user memory and session logs — something you should consciously permit because it can expose sensitive information.

Purpose & Capability

Name/description match the runtime instructions: the skill's goal is to review recent activity and detect cognitive biases, and the SKILL.md directs the agent to gather recent decisions/tasks and apply mental models. Access to memory and session logs is coherent with that purpose.

Instruction Scope

Instructions tell the agent to read today's memory file (memory/YYYY-MM-DD.md) and to scan 'session logs' for today's activity. That behavior is expected for a review task, but the skill does not declare those config paths in its metadata and 'session logs' is unspecified — the agent could access different log locations depending on implementation. This is a privacy scope note rather than a clear mismatch in intent.

Install Mechanism

There is no install spec or code — the skill is instruction-only, so nothing is downloaded or written to disk. This minimizes supply-chain risk.

Credentials

Skill declares no environment variables, credentials, or config paths. The runtime instructions reference local files (memory and logs) but do not request cloud keys or unrelated secrets. The requested access is proportionate for the stated purpose.

Persistence & Privilege

always is false and the skill is user-invocable; it does not request persistent/system-wide privileges. Scheduling is optional and uses a suggested trigger message (MUNGER_OBSERVER_RUN), which is a benign convenience but should be configured deliberately.

Assessment

The skill appears internally consistent, but it will read your local memory file and session logs to do reviews. Before enabling automated runs, confirm where those files live and what they contain (they may include sensitive text, PII, or secrets). Prefer manual invocation initially, inspect a sample report, and if you enable scheduling, restrict the skill's file-access scope or retention. If you want stronger guarantees, ask the author to declare the exact config paths the skill will read (or modify the SKILL.md to limit log sources) so you can audit access.

Automated daily review applying Charlie Munger's mental models to surface blind spots and cognitive traps.

Process

1. Gather Today's Activity

• Read today's memory file (memory/YYYY-MM-DD.md)
• Scan session logs for today's activity
• Extract: decisions made, tasks worked on, problems tackled, user requests

2. Apply Mental Models

Inversion

• What could go wrong? What's the opposite of success here?
• "Tell me where I'm going to die, so I'll never go there."

Second-Order Thinking

• And then what? Consequences of the consequences?
• Short-term gains creating long-term problems?

Incentive Analysis

• What behaviors are being rewarded? Hidden incentive structures?
• "Show me the incentive and I'll show you the outcome."

Opportunity Cost

• What's NOT being done? Cost of this focus?
• Best alternative foregone?

Bias Detection

• Confirmation bias: Only seeking validating information?
• Sunk cost fallacy: Continuing because of past investment?
• Social proof: Doing it because others do?
• Availability bias: Overweighting recent/vivid information?

Circle of Competence

• Operating within known territory or outside?
• If outside, appropriate humility/caution?

Margin of Safety

• What's the buffer if things go wrong?
• Cutting it too close anywhere?

3. Generate Output

If insights found: 1-2 concise Munger-style observations If nothing notable: "All clear — no cognitive landmines detected today."

Output Format

🧠 **Munger Observer** — [Date]

[Insight 1: Model applied + observation + implication]

[Insight 2 if applicable]

— "Invert, always invert." — Carl Jacobi (Munger's favorite)

Example

🧠 **Munger Observer** — January 19, 2026

**Opportunity Cost Alert:** Heavy focus on infrastructure today. The content queue is aging — are drafts decaying in value while we polish tools?

**Second-Order Check:** Speed improvement is good first-order thinking. Second-order: faster responses may raise expectations for response quality. Speed without substance is a trap.

— "Invert, always invert."

Scheduling (Optional)

Set up a cron job for daily automated review:

• Recommended time: End of workday (e.g., 5pm local)
• Trigger message: MUNGER_OBSERVER_RUN