メニュー
Design metacognitive checkpoints that prevent AI-assisted learning from bypassing genuine understanding. Use when students use AI tools and may overestimate their own comprehension.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
SOC 職業分類に基づく
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/GarethManning/education-agent-skills --skill metacognitive-monitoring-ai-contextsコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
After scaffolded practice, run an unassisted check with no AI help. Use to distinguish supported performance from what the learner can currently do independently.
Review the week using accumulated session evidence — retrieval rates, hint depths, calibration accuracy, transfer and unassisted results. The learner identifies patterns and sets a strategy goal. Use weekly or after a multi-session period.
Design discipline-specific critical thinking tasks grounded in knowledge-contingent reasoning rather than generic skills. Use when embedding higher-order thinking into subject content.
| name | metacognitive-monitoring-ai-contexts |
| description | Design metacognitive checkpoints that prevent AI-assisted learning from bypassing genuine understanding. Use when students use AI tools and may overestimate their own comprehension. |
| disable-model-invocation | false |
| user-invocable | true |
| effort | medium |
| skill_id | ai-learning-science/metacognitive-monitoring-ai-contexts |
| skill_name | Metacognitive Monitoring in AI Contexts |
| domain | ai-learning-science |
| version | 1.0 |
| evidence_strength | moderate |
| evidence_sources | ["Thiede et al. (2003) — Summarizing can improve metacomprehension accuracy","Winne & Hadwin (1998) — Studying as self-regulated learning (SRL model)","Dunning et al. (2003) — Why people fail to recognize their own incompetence (Dunning-Kruger)","Bjork et al. (2013) — Self-regulated learning: beliefs, techniques, and illusions","Kazemitabaar et al. (2023) — Studying the effect of AI code generators on supporting novice learners in introductory programming"] |
| input_schema | {"required":[{"field":"ai_learning_context","type":"string","description":"The specific context in which students are using AI tools for learning — what they are doing with AI and what they are supposed to be learning"},{"field":"metacognitive_risk","type":"string","description":"The specific metacognitive risk to address — how AI use might distort students' self-assessment of their own understanding"}],"optional":[{"field":"student_level","type":"string","description":"Age/year group and proficiency level"},{"field":"subject_area","type":"string","description":"The curriculum subject"},{"field":"ai_tool","type":"string","description":"Which AI tool students are using — ChatGPT, Copilot, a custom tutoring system, or other"},{"field":"assessment_context","type":"string","description":"How student learning will be assessed — exam, project, practical demonstration, or other"}]} |
| output_schema | {"type":"object","fields":[{"field":"metacognitive_diagnosis","type":"object","description":"Analysis of how AI use in this context might distort metacognitive monitoring — the specific risks and mechanisms"},{"field":"monitoring_interventions","type":"array","description":"Specific strategies to improve metacognitive accuracy — prompts, calibration tasks, and self-assessment tools"},{"field":"ai_usage_guidelines","type":"object","description":"When and how to use AI in this context to support rather than undermine metacognition"},{"field":"assessment_alignment","type":"object","description":"How to align assessment with metacognitive goals — testing what students actually know, not what the AI knows"}]} |
| chains_well_with | ["self-explanation-prompt-designer","ai-feedback-design-principles","productive-failure-desirable-difficulty-designer","formative-assessment-loop-designer","ai-learning-boundary-mapper","ai-output-critical-audit-designer"] |
| teacher_time | 4 minutes |
| tags | ["metacognition","self-regulation","Dunning-Kruger","overconfidence","AI-literacy","Winne","Thiede","calibration"] |
Analyses how AI tool use in a specific learning context might distort students' metacognitive monitoring — their ability to accurately assess what they know and don't know — and designs interventions to maintain metacognitive accuracy. This is one of the most urgent challenges in AI-enabled education. When a student uses an AI tool to complete work, they may experience a fluency illusion: the work looks good, the answers are correct, the text is fluent — and the student concludes "I understand this." But the STUDENT didn't do the cognitive work; the AI did. The student's sense of understanding is calibrated to the PRODUCT (which is good) rather than to their OWN knowledge (which may be unchanged). Bjork et al. (2013) showed that learners are systematically poor at judging their own learning — they confuse familiarity with understanding, and fluent performance with durable knowledge. AI tools dramatically amplify this miscalibration because they produce fluent, correct output that the student may mistake for evidence of their own competence. The output includes a metacognitive diagnosis (how AI use distorts self-assessment in this specific context), monitoring interventions (strategies to improve metacognitive accuracy), AI usage guidelines (when to use and when to restrict AI), and assessment alignment (ensuring tests measure student knowledge, not AI-assisted performance).
Winne & Hadwin (1998) developed the most comprehensive model of self-regulated learning (SRL), which places metacognitive monitoring at its centre. Their model describes a cycle: the learner sets goals, applies strategies, monitors whether the strategies are working, and adjusts. Effective learning depends critically on the MONITORING stage — the learner's ability to accurately judge whether they are understanding the material. When monitoring is inaccurate (the learner thinks they understand when they don't), the entire self-regulation cycle breaks down: they stop studying too early, choose inappropriate strategies, and are surprised by poor assessment results. Thiede et al. (2003) showed that metacomprehension accuracy (the correlation between judged and actual understanding) is typically very low — around r = 0.27. However, they found that certain activities dramatically improve accuracy: delayed summary writing, keyword generation, and any task that forces the learner to generate from memory rather than recognise from the text. The key principle: metacognitive accuracy improves when the monitoring task requires RETRIEVAL, not just recognition. Dunning et al. (2003) documented the Dunning-Kruger effect: the least competent individuals are the MOST overconfident in their abilities, because they lack the knowledge needed to recognise their own incompetence. In AI contexts, this effect may be amplified: a student who doesn't understand a concept cannot distinguish their own (poor) understanding from the AI's (excellent) output. Bjork et al. (2013) reviewed the psychology of self-regulated learning and identified several "illusions of competence" — conditions where learners feel they've learned more than they actually have. These include: familiarity (having seen something before feels like understanding it), fluency (material that's easy to process feels like it's well-learned), and performance (doing well now feels like permanent learning). AI tools can trigger all three illusions simultaneously: the AI-produced output is familiar (the student saw it being generated), fluent (LLMs produce polished text), and high-performing (the answers are correct). Kazemitabaar et al. (2023) studied how AI code generators (like Copilot) affect novice programming learners and found that while AI-assisted students completed tasks faster and with fewer errors, they showed weaker understanding on subsequent tasks without AI support. The students had learned to use the AI, not to program. This is a direct empirical demonstration of the metacognitive risk: AI assistance produced the ILLUSION of learning without the REALITY of learning.
The teacher must provide:
Optional (injected by context engine if available):
You are an expert in metacognition and self-regulated learning, with deep knowledge of Winne & Hadwin's (1998) SRL model, Thiede et al.'s (2003) metacomprehension accuracy research, Dunning et al.'s (2003) work on the Dunning-Kruger effect, Bjork et al.'s (2013) illusions of competence, and emerging research on AI's impact on metacognition (Kazemitabaar et al., 2023). You understand that AI tools pose a specific and novel threat to metacognitive monitoring: they produce fluent, correct output that students mistake for evidence of their own understanding. This is not a minor concern — it is potentially the most significant educational risk of AI tools, because it undermines the self-regulation cycle that drives all independent learning.
CRITICAL PRINCIPLES:
- **The core problem is CALIBRATION.** Metacognitive monitoring works when students' confidence matches their competence. AI distorts calibration by inflating confidence (the work looks great) without necessarily increasing competence (the student may not have learned anything). The interventions must improve calibration, not just raise or lower confidence.
- **Fluency ≠ understanding.** When AI produces smooth, well-structured output, students experience processing fluency — the content feels easy to understand. But ease of processing does not indicate depth of learning. In fact, Bjork et al. (2013) showed that material that is HARDER to process (disfluent fonts, challenging language, interleaved examples) often produces BETTER learning. AI removes this desirable difficulty.
- **The solution is not banning AI.** It's redesigning the learning process so that students ENCOUNTER THEIR OWN KNOWLEDGE STATE — not just the AI's output. This means creating moments where students must produce from memory, without AI support, and compare their production to what they thought they knew.
- **Retrieval-based monitoring is the gold standard.** Thiede et al. (2003): the most effective way to improve metacognitive accuracy is to require RETRIEVAL — generating from memory rather than recognising from presented material. After using AI, students should close the AI, attempt the task from memory, and compare. This reveals the gap between perceived and actual understanding.
- **Metacognitive monitoring must be DESIGNED IN, not added on.** If you wait until the assessment to discover that students thought they knew the material but didn't, it's too late. Monitoring checkpoints must be built into the learning process — at the point of AI use, not after it.
Your task is to analyse the metacognitive risks and design monitoring interventions for:
**AI learning context:** {{ai_learning_context}}
**Metacognitive risk:** {{metacognitive_risk}}
The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."
**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the context.
**AI tool:** {{ai_tool}} — if not provided, assume a general-purpose LLM chatbot.
**Assessment context:** {{assessment_context}} — if not provided, assume a traditional exam without AI access.
Return your output in this exact format:
## Metacognitive Monitoring Analysis: [Context Description]
**Context:** [How students are using AI]
**Core risk:** [The specific metacognitive distortion — one sentence]
**Severity:** [How likely and how damaging this risk is — high/moderate/low]
### Metacognitive Diagnosis
[Detailed analysis of how AI use in this context distorts metacognitive monitoring. Name the specific illusions of competence at play. Explain the mechanism — HOW does the AI use lead to miscalibrated confidence?]
### Monitoring Interventions
[Specific strategies to improve metacognitive accuracy. For each:]
**Intervention [N]: [Name]**
- **What:** [What the student does]
- **When:** [At what point in the learning process — before, during, or after AI use]
- **Why it works:** [The metacognitive principle it applies]
- **Example:** [A concrete example of the intervention in this context]
### AI Usage Guidelines
[When to use AI and when to restrict it — specific, practical guidelines for this context]
### Assessment Alignment
[How to design assessment so that it measures student knowledge, not AI-assisted performance]
### Red Flags
[Observable signs that metacognitive distortion is occurring — what the teacher should watch for]
**Self-check before returning output:** Verify that (a) the diagnosis identifies the specific metacognitive mechanism, (b) interventions target calibration not just confidence, (c) retrieval-based monitoring is included, (d) AI is not banned but strategically managed, and (e) assessment aligns with metacognitive goals.
Scenario: AI learning context: "Year 12 A-level English Literature students use ChatGPT to help write essays. The typical workflow: student receives an essay question (e.g., 'How does Shakespeare present the theme of power in Macbeth?'), enters it into ChatGPT, receives a well-structured essay with relevant quotes and analysis, then edits and personalises the AI output. Most students change some wording and add a few of their own points, but the core argument and structure are AI-generated." / Metacognitive risk: "Students believe they 'understand' the literary analysis because the essay looks good and they can follow the argument when reading it. But when asked to discuss the text in a class seminar without AI, or in a closed-book exam, they struggle to articulate the argument, recall specific quotes, or build an analytical case from scratch. They are surprised by this — they thought they understood it." / Student level: "Year 12, A-level English Literature" / AI tool: "ChatGPT (free version)" / Assessment context: "Closed-book A-level exam — no AI access, 45 minutes per essay"
Context: Year 12 students use ChatGPT to generate essay drafts, then edit and personalise them Core risk: Students confuse their ability to RECOGNISE good literary analysis (when reading AI output) with their ability to PRODUCE it (from their own knowledge) Severity: High — the gap between AI-assisted performance and unassisted exam performance could be substantial, and students are likely to discover this gap only at the point of assessment
Three illusions of competence are operating simultaneously:
1. Recognition-production confusion (the most damaging). When a student reads the AI's analysis of power in Macbeth, they follow the argument. It makes sense. They nod along. They may even think "Yes, that's what I thought." But FOLLOWING an argument and CONSTRUCTING an argument are completely different cognitive tasks. Reading the AI's essay activates recognition ("I understand this when I see it") but does not build production capacity ("I can generate this myself"). The student's metacognitive monitoring is calibrated to their recognition ability, not their production ability — and the exam tests production.
2. Fluency illusion. ChatGPT produces polished, well-structured prose. When students read and edit this output, they experience processing fluency — it feels easy to understand, and therefore feels well-learned. Bjork et al. (2013) showed that fluency is one of the most misleading cues for learning. The student's confidence is high because the material feels smooth, not because their understanding is deep.
3. Effort misattribution. Students spend time editing the AI output — changing words, adding points, reorganising paragraphs. This feels like cognitive work, and students may attribute their sense of understanding to this editing effort. But editing someone else's analysis is a fundamentally different cognitive task from constructing your own. The editing effort produces familiarity with THIS specific essay, not transferable analytical skill.
The Dunning-Kruger amplifier: Students who are weakest in literary analysis benefit most from the AI (the gap between their ability and the AI's is largest) and are LEAST able to recognise the gap. A strong student who reads the AI's essay can identify where it differs from their own thinking. A weak student reads the same essay and experiences only agreement — they can't see what they're missing because they don't have the knowledge to see it.
Intervention 1: The "Close the Laptop" Test
Intervention 2: Quote Recall Before AI Use
Intervention 3: Explain-It-Back Protocol
Intervention 4: Prediction-Outcome Calibration
AI is helpful for:
AI is harmful when:
Recommended workflow:
The assessment must test what the STUDENT knows, not what the AI-student system produces. For A-level English Literature:
Closed-book timed essays are already well-aligned — the student must produce from memory. Ensure students have practised this format regularly, not just AI-assisted homework essays.
Class seminars and discussions are excellent metacognitive tests — verbal analysis in real time cannot be AI-assisted. Use these as formative assessments throughout the term.
Quote tests (write 10 relevant quotes from memory for this theme) reveal knowledge of the text separately from analytical skill.
"Think aloud" analysis — give the student a passage they haven't seen and ask them to analyse it live, talking through their thinking. This tests the analytical SKILL, not the product.
Watch for these signs that metacognitive distortion is occurring:
The evidence on AI-specific metacognitive effects is still emerging. Kazemitabaar et al. (2023) is one of a small but growing number of studies on AI tools and metacognition. The broader metacognitive research (Thiede et al., 2003; Bjork et al., 2013) provides strong theoretical grounding, but the specific application to LLM-assisted learning is based on extrapolation from these principles, not extensive empirical testing.
Monitoring interventions add cognitive and time costs. The "close the laptop" test, quote recall, and explain-it-back protocols all require additional time and effort. In contexts where students are under time pressure (heavy workloads, multiple subjects), adding metacognitive monitoring exercises may feel burdensome. Teachers must balance metacognitive accuracy against practical feasibility.
Individual differences in metacognitive ability are large. Some students are naturally good at monitoring their own understanding; others are not. The Dunning-Kruger effect suggests that the students who most need metacognitive support are the least likely to recognise that they need it. Interventions must be STRUCTURAL (built into the workflow for all students) rather than ADVISORY ("you should check your understanding").
The relationship between AI use and metacognition may be more nuanced than "AI harms metacognition." Some uses of AI (e.g., using AI to generate practice questions, then attempting them without AI) might actually IMPROVE metacognitive accuracy by creating retrieval opportunities. The risk is context-dependent, not absolute. The diagnosis above applies specifically to the "AI generates, student edits" workflow — other workflows may have different metacognitive profiles.