Skylv Metacognition Engine

Enables AI agents to reflect on their own reasoning, detect cognitive biases, and improve decision quality through structured self-examination loops.

Security Scan

The skill's code and instructions align with the stated purpose (metacognition/bias detection); it is a local, instruction-driven JS helper that reads files only when asked and does not request credentials or network access.

Purpose & Capability

Name/description match the actual artifacts: SKILL.md describes metacognitive checks and the bundled metacognition_engine.js implements text-based bias detection and reflection. Required resources (none) are proportionate to the purpose.

Instruction Scope

SKILL.md keeps scope to self-reflection steps and bias checks. The included JS offers commands to analyze a local file (fs.readFileSync) and reflect on provided reasoning strings — this is consistent with the skill but means the tool will read any file path passed to its analyze command. It does not instruct broad or silent collection of system state or credentials.

Install Mechanism

No install spec (instruction-only) and only a small JS file is bundled. Nothing is downloaded or executed from external URLs. Risk from install mechanism is low.

Credentials

The skill does not request environment variables, credentials, or config paths. There are no apparent requests for unrelated secrets or external service tokens.

Persistence & Privilege

always is false and the skill does not request persistent/privileged agent-wide changes. It does not modify other skills or system configuration. Autonomous invocation is allowed by default but is not combined with broad credential or network access here.

Assessment

This skill is internally consistent with its metacognition purpose and appears low-risk: it implements simple regex-based bias detectors and an explicit command to analyze local files. Before installing, consider: 1) If your agent has filesystem access, the 'analyze' command will read any file path you or the agent supplies — avoid giving it access to sensitive files. 2) The detection logic is heuristic and may produce false positives/negatives (regex-based); treat suggestions as guidance, not authoritative diagnosis. 3) There is no network or credential usage in the code, so the main risk is accidental local-file exposure rather than exfiltration. If you want extra assurance, review the metacognition_engine.js file yourself or run it in a sandboxed environment.

Metacognition Engine

Give your AI agent the ability to think about its own thinking.

What is Metacognition?

Metacognition = "thinking about thinking." This skill enables AI agents to:

• Detect when they're uncertain or confused
• Identify reasoning gaps before they cause errors
• Recognize cognitive biases in their own output
• Self-correct before delivering answers

Core Framework

1. Pre-Output Check

Before responding, run through these questions:

1. Am I confident in this answer? (Yes / Partial / No)
2. What are the 3 most likely ways this could be wrong?
3. What information would I need to be 100% certain?

2. Cognitive Bias Detection

Check for common biases:

• Anthropomorphism — projecting human traits onto AI
• Authority bias — deferring to stated credentials without verification
• Hindsight bias — acting like something was obvious after the fact
• Confirmation bias — seeking only confirming evidence

3. Uncertainty Quantification

Express confidence explicitly:

Confidence	Meaning	Action
90%+	Highly confident	Answer directly
70-89%	Likely correct	Answer + add caveat
50-69%	Uncertain	Ask clarifying questions
<50%	Likely wrong	Decline or escalate

Example

Without metacognition:

"The capital of France is Paris."

With metacognition:

"Based on my training data, the capital of France is Paris (confidence: 95%). Note: My knowledge has a cutoff date. For real-time data, verify current information."

Use Cases

• Critical decisions: Add metacognition checkpoint before any consequential answer
• User corrections: When a user corrects you, analyze WHY you were wrong
• Complex problems: Run bias detection before solving multi-step problems
• Knowledge boundaries: Automatically flag when you're approaching your knowledge limit

MIT License © SKY-lv