m3taversal
be8ff41bfe
Wrote sourced_from: into 414 claim files pointing back to their origin source. Backfilled claims_extracted: into 252 source files that were processed but missing this field. Matching uses author+title overlap against claim source: field, validated against 296 known-good pairs from existing claims_extracted. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
41 lines
2.1 KiB
Markdown
41 lines
2.1 KiB
Markdown
---
|
|
type: source
|
|
title: "25 Prompts for Making AI Agents Self-Diagnose"
|
|
author: "kloss (@kloss_xyz)"
|
|
url: https://x.com/kloss_xyz/status/2032223154094162063
|
|
date_published: 2026-03-09
|
|
date_archived: 2026-03-16
|
|
domain: ai-alignment
|
|
status: processed
|
|
processed_by: theseus
|
|
tags: [agent-self-diagnosis, metacognition, oversight-scaffolding, prompt-engineering]
|
|
sourced_via: "Leo routed from X ingestion (@kloss_xyz tweet 2032223154094162063)"
|
|
claims_extracted:
|
|
- "structured self-diagnosis prompts induce metacognitive monitoring in AI agents that default behavior does not produce because explicit uncertainty flagging and failure mode enumeration activate deliberate reasoning patterns"
|
|
---
|
|
|
|
# 25 Prompts for Making AI Agents Self-Diagnose
|
|
|
|
Practitioner-generated prompt collection for inducing metacognitive monitoring in AI agents. Published as a tweet thread by @kloss_xyz.
|
|
|
|
## Prompt Categories (my analysis)
|
|
|
|
**Uncertainty calibration (5):** #4 confidence rating, #5 missing information, #15 evidence quality, #16 deductive vs speculative, #23 likely→certain threshold
|
|
|
|
**Failure mode anticipation (4):** #1 biggest failure risk, #6 what wrong looks like, #11 three most likely failure modes, #19 what context invalidates approach
|
|
|
|
**Tool/output verification (3):** #2 schema verification, #7 expected tool return, #8 actual vs expected comparison
|
|
|
|
**Strategy meta-monitoring (4):** #9 step count check, #13 redo from scratch, #18 solving right problem, #20 loop detection
|
|
|
|
**Adversarial self-review (3):** #12 argue against answer, #14 expert critique, #17 simplest explanation (Occam's)
|
|
|
|
**User alignment (3):** #10 unstated user intent, #21 define done, #25 optimize for user's use case
|
|
|
|
**Epistemic discipline (3):** #22 replace "I think" with evidence, #24 simpler solution check, #3 flag uncertainty explicitly
|
|
|
|
## Evidence Base
|
|
No empirical validation of these prompts. This is practitioner knowledge, not a study. However, connects to validated finding that structured prompting produces measurable performance gains (Residue prompt reduced human intervention 6x — Reitbauer 2026).
|
|
|
|
## Extraction Status
|
|
- 1 claim: structured self-diagnosis prompting as oversight scaffolding
|