Skills DirectorySkills Directory
SkillsLearnSecurityCategoriesDocsCommunityBlog
Sign InSubmit Skill
Skills Directory

Security-tested agent skills for Claude, coding agents, and AI workflows.

Directory

  • Browse Skills
  • All Skills A–Z
  • Claude Skills
  • Claude Code Skills
  • Agent Skills
  • Categories
  • Submit a Skill

Learn

  • Learn Hub
  • Install Claude Skills
  • Write SKILL.md
  • Skills vs MCP
  • Directories Compared

Security

  • Security
  • Methodology
  • Secure Claude Skills
  • Security Badges

Company

  • About
  • Community
  • Blog
  • API Docs
  • Advertise

2026 Skills Directory. All rights reserved.

Back to skills

Skill Training

ASecurity

Update a skill from how a real run actually went. Use right after a skill produces output the user corrects, rejects, or edits before using. Also use when the user says a skill keeps making the same mistake, or asks to make something remember a preference. This is the loop that makes skills improve instead of drifting.

7 stars
0 votes
0 copies
0 views
Added 9/22/2026
ai-agentsgo

Security Analysis

A100/100

Scanned 9/22/2026

Install to Claude Code

$npx -y skills add calebnewtonusc/Chewbacca --skill skill-training --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Skill Training?

Add the live security badge to your README — it updates automatically with every re-scan.

Security grade badge for Skill Training
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/calebnewtonusc-skill-training/badge)](https://www.skillsdirectory.com/skills/calebnewtonusc-skill-training)

More formats (shields.io, HTML) on the badges page.

Download Zip
Files
SKILL.md
---
name: skill-training
description: Update a skill from how a real run actually went. Use right after a skill produces output the user corrects, rejects, or edits before using. Also use when the user says a skill keeps making the same mistake, or asks to make something remember a preference. This is the loop that makes skills improve instead of drifting.
---

# skill-training

A skill that is never updated from real runs decays: the world moves and the
file does not. The correction the user just made is the training data, and it
is gone the moment this session closes.

## When this fires

The user edited the output before using it. Or rejected it. Or said "not like
that." Or fixed the same thing twice. That is the signal, and it is easy to
miss because the work still got done.

## The loop

1. **Name the delta.** What did the output say, what did the user change it
   to, and what is the difference in one sentence. Not the diff. The rule
   behind the diff.
2. **Ask which kind it is.** Three answers, and the user picks:
   - **Persist.** Always do it this way from now on.
   - **One-off.** This case was special, change nothing.
   - **Always ask.** Put a question in the skill at this point.
   Guessing here is how a skill acquires rules nobody wanted.
3. **Edit the skill file**, not the output. Fixing the artifact solves today.
   Fixing the file solves the next twenty.
4. **Append a dated changelog line** to the skill saying what changed and
   what prompted it. A rule whose reason is lost gets removed by whoever
   finds it unconvincing.
5. **Verify the edit landed.** Read the file back and confirm the new text is
   there. A stale string match fails silently, and a change you reported but
   never applied is worse than no change, because it stops both of you from
   ever looking at that spot again.

## Anchor the rule to its incident

A constraint carrying only a rationale gets raised by the next person who
disagrees with the rationale. A constraint carrying what actually happened
does not. Write the observation, not the reasoning.

## Where the lesson goes

Not everything belongs in the skill.

- **A reusable judgment** goes in the skill, or in a rule under
  `~/.claude/rules/` when it spans skills.
- **A repeatable mechanical failure** becomes a test next to the code. A
  paragraph does not fail CI.
- **A fact about the user, a project or a person** goes in the second brain,
  not in the skill. Skills hold instructions. Files hold facts.

If a correction produces none of those, it was a typo. Fix it and move on.

## Keep the skill short

Every training pass wants to add a line. Length is what stops a skill being
obeyed, so a pass that adds should also look for what the addition makes
redundant.

Attribution

calebnewtonusccalebnewtonusc
View sourceMore from calebnewtonusc →
SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments (0)

No comments yet. Be the first to comment!

SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Related Skills

Caveman

Ultra-compressed communication mode that cuts output tokens while keeping technical accuracy. Levels: lite, full, ultra and the wenyan variants. Use for /caveman, "caveman mode", "talk like caveman", "be brief" or "less tokens".

1066601 votes

Hyperplan

Adversarial multi-agent planning skill. Self-orchestrates 5 hostile category members (unspecified-low, unspecified-high, deep, ultrabrain, artistry) via team-mode for ruthless cross-critique debate, distills only the defensible insights, then MANDATORILY hands the distilled insight bundle to the `plan` agent for executable plan formalization. Use when planning needs maximum rigor and surfacing of weak assumptions, blind spots, and over-engineering. Triggers: 'hyperplan', 'hpp', '/hyperplan', ...

686011 votes

Mcp Code Execution

Routes multi-tool workflows through MCP servers for large datasets and pipelines. Use when Bash tool overhead is limiting throughput on data-heavy tasks.

3351 votes

catchup

Recovers the conversation and failed tool calls of a previous Codex, Claude Code, Antigravity, Cline, Copilot CLI, Cursor, DeepSeek Harness, Kimi, OpenCode, Pi Agent, or ZCode session. Use when the user says "catch up", "what did the last session do", "get me up to speed", "I switched agents", asks to recover/summarize a previous session before continuing, or asks to diagnose or report a catchup failure. Do NOT use for the current conversation, git history, or any non-agent log.

651 votes

math-skill

A comprehensive mathematical reasoning skill for AI assistants — handles arithmetic to research-level problems with rigorous step-by-step reasoning, systematic verification, and transparent uncertainty handling

381 votes
View all in ai-agents →