Skip to main content
prompt GP ChatGPT

meta_skill — Claude Code prompt: what it does, how to install

The meta_skill Claude Code prompt guides agents through finding existing skills and recording lessons backed by a task. It comes into play when an agent needs to add a lesson or revise a skill.

MIT ★ 488 updated 2026-11-15

Install into Claude Code · Manually: put the files in place All methods and platforms ↓
.claude/commands/meta_skill.md
Open on GitHub

What meta_skill does

The prompt sets out a search-and-write workflow. The agent lists skill files with list_skills, reads a relevant one with read_skill, then decides whether the lesson belongs in a new file or an existing skill.

A lesson must come from evidence in the current attempt, such as a failure mechanism or a successful recovery. Text passed to write_skill captures the task, outcome, conditions, attempts, evidence and limits on reuse; hypotheses stay separate from verified observations. The shared general/ directory is read-only, while writing is limited to the current namespace.

The prompt rules out turning one task’s result into a universal rule or storing secrets and large logs. Recording experience does not establish the current state or replace verification, recovery, or an honest account of partial attempts.

What it helps with

  • find relevant skills before writing a new lesson
  • revise an existing skill while retaining useful guidance and evidence
  • separate confirmed observations from hypotheses in a skill

Example prompts for the agent

  • Find relevant skills, then save the confirmed lesson from this task.
  • Read the existing skill and update it with a lesson supported by this attempt.
  • Record the failure and recovery, keeping observed facts separate from hypotheses.

How to install meta_skill

Manually: put the files in place

.claude/commands/meta_skill.md

Questions about meta_skill

How do I install meta_skill in Claude Code?

Place meta_skill.md in .claude/commands/.

What does meta_skill do?

It guides an agent through locating existing skills and saving lessons supported by task evidence. It also sets boundaries for where skills can be written and what the record should contain.

What kind of lessons can meta_skill record?

Only lessons supported by the current attempt, such as an observed cause of failure or a successful recovery. Hypotheses must be distinguished from verified observations.

Source text

meta_skill.md · MIT
Skill writing is enabled for this session. First discover existing guidance
with list_skills without arguments. This lists Markdown paths in general/ and
the current environment namespace, including operation skills and earlier experience,
without loading contents. Pass a relevant path to read_skill before
deciding whether a reusable lesson is missing or an existing skill should be
updated. Reuse a listing already in context unless the directory may have changed.

Use write_skill only for a lesson supported by this attempt: an observed failure
mechanism, successful recovery, decision-changing evidence check or condition
under which an approach is inappropriate. Do not create a skill to satisfy a
quota or present an untested idea as a verified technique.

Discovery does not grant writing permission. Follow write_skill's exposed path
policy; this session permits Markdown skills under
`{{skill_namespace}}/<topic>.md`, including ordinary skill subdirectories.
List, read and write use the same configured skills library. The shared general/
directory is read-only, even when it provides relevant guidance. Keep supported
environment-specific lessons in the writable namespace; do not copy the shared
library or promote one task's outcome to a universal rule. Do not modify other
skill library namespaces, prompts or implementation files. Before updating an existing skill,
read it and preserve relevant prior guidance and evidence. Choose a distinct
concise topic when the experience differs.

Pass the full Markdown text to write_skill. Record the task context, observed
outcome, relevant conditions, attempts, evidence and limits on reuse. Separate
verified observations from hypotheses. Prefer qualitative geometry and decision
rules to an action replay. Do not store scene coordinates, object identities or
gripper conventions as universal constants; later runs must use their own
calibration, observations and tool schemas. Do not store secrets or large logs.

Preserve the unfinished task and its physical decision boundaries while writing.
Experience does not establish current state or replace verification, recovery
or finish. Report failed and partial attempts honestly.

© Open-RUA · MIT · Open-RUA/uni-robot-agent

Similar prompts

All prompts for Writing and docs →

From the same set41

The whole set →
return_background The return_background Claude Code prompt coordinates background jobs and waits. It defines queue boundaries, result checks, and recovery after failures. GP ChatGPT ★ 488repo stars main_agent This Claude Code prompt delegates visual observation while leaving physical actions and task completion to the main agent. The main agent checks observation evidence before making a decision that depends on it. GP ChatGPT ★ 488repo stars hybrid The hybrid Claude Code prompt helps choose between actor_act and direct robot motion. It guides a bounded control phase and the checks that follow. GP ChatGPT ★ 488repo stars hybrid The hybrid prompt for Claude Code and ChatGPT guides a Go2 quadruped through a prepared indoor navigation episode. The agent chooses movement from RGB views and ends the run only by calling finish. GP ChatGPT ★ 488repo stars observer The observer Claude Code prompt inspects published observations for an assigned visual question. It reports visible evidence and flags uncertainty or a useful next observation. GP ChatGPT ★ 488repo stars common The common prompt for Claude Code and ChatGPT guides agents through physical robot tasks using observations and environment instructions. It sets checks for state changes and recovery after uncertain execution. GP ChatGPT ★ 488repo stars hybrid_decision The hybrid_decision prompt for Claude Code reviews an agent’s proposed action against observations and execution history. It selects whether to continue, stop, or use an allowed correction, then sends one submit_review call. GP ChatGPT ★ 488repo stars code_as_policy The code_as_policy Claude Code prompt defines how an agent controls a robot through its generated Robot API. The agent runs Python programs, checks tool results and inspects images before dependent actions. GP ChatGPT ★ 488repo stars

Project sponsors

Companies that keep this analytics open