
from dots9
Absorb external signals (articles, observations, ideas, failures) into the global agent harness: extract insights, map gaps, apply fixes, and measure absorption
Meta-Prompt is a meta-skill that ingests signals—external articles, user observations, ideas, or agent failure reports—and runs a structured loop to evolve the global agent harness (~/.claude). It extracts actionable insights, compares them against existing rules and skills, proposes gap-filling changes, and measures an 'absorption rate' for applied improvements. The skill is procedural and prescriptive, designed to keep global behavior aligned with real-world inputs.
Use when you have an external reference (URL), a systemic problem across skills, or ideas for harness-wide improvements. Do NOT use this for single-project harness changes; delegate those to a project-level harness-engineering workflow.
.claude/meta-prompt/), though no external scripts are required to run the loopDesigned for agents that can read repository SKILL.md files and edit harness files or create tasks for subordinate skills (harness-engineering, my-skill-creator). Works well where agent tooling supports WebFetch, recall, and file edits.
Meta-prompt skill for absorbing external signals (articles, observations, ideas, failures) into a global Claude Code harness (~/.claude/). Structured as a 5-phase process (triage, extract, compare, apply, measure) with user confirmation gates. Entirely in Korean, limiting accessibility. No bundled scripts — purely prompt-driven. Targets a very specific harness setup (ssiumha/dots repo).
Well-structured process skill with clear phases and user confirmation gates (AskUserQuestion). No security concerns — no code execution, no network calls, no file operations beyond standard agent tools. Main limitations are language barrier (Korean-only) and specificity to one user's harness setup.