Skip to content
Back to skills

Agents Md Improver

ASecurity

Proposes updates to AGENTS.md so repo-local coding-agent instructions stay accurate, non-contradictory, and consolidated. Use when the user corrects the agent (e.g. do not do X, always do Y, from now on, remember this), asks to edit or sync AGENTS.md, CLAUDE.md, .claude/CLAUDE.md, or GEMINI.md, wants one canonical instruction file or to deduplicate agent docs, or flags conflicting or outdated repository agent rules.

  • 39 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 26, 2026
ai-agents

Security analysis

A100/100

Scanned May 27, 2026

npx -y skills add ericmjl/skills --skill agents-md-improver --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Agents Md Improver?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Agents Md Improver
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ericmjl-agents-md-improver/badge)](https://www.skillsdirectory.com/skills/ericmjl-agents-md-improver)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: agents-md-improver
description: Proposes updates to AGENTS.md so repo-local coding-agent instructions stay accurate, non-contradictory, and consolidated. Use when the user corrects the agent (e.g. do not do X, always do Y, from now on, remember this), asks to edit or sync AGENTS.md, CLAUDE.md, .claude/CLAUDE.md, or GEMINI.md, wants one canonical instruction file or to deduplicate agent docs, or flags conflicting or outdated repository agent rules.
license: MIT
---

# AGENTS.md improver

This skill helps keep repository instructions accurate and consistent. When a user corrects the coding agent's behavior, treat it as a potential update to the repo's instruction files and propose a clean, integrated edit to `AGENTS.md`.

## Usage

Use this skill anytime the user:

- Corrects the agent ("don't do X", "do Y instead", "from now on...", "remember to...", "add a rule that...").
- Asks to update instruction files like `AGENTS.md`, `CLAUDE.md`, `.claude/CLAUDE.md`, or `GEMINI.md`, or to align or consolidate those files.
- Notices contradictions, duplication, or drift between repository instruction files.

## Requirements

- Repo-local file access to read and edit `AGENTS.md`.
- If present in the repo, read access to `CLAUDE.md` and/or `.claude/CLAUDE.md`.

## What It Does

- Proposes a patch to incorporate the user's correction into `AGENTS.md` seamlessly.
- Looks for contradictions and resolves them in the proposal (or asks for clarification when needed).
- If repo-local `CLAUDE.md` or `.claude/CLAUDE.md` exists, proposes consolidating instructions into `AGENTS.md`, then offers follow-up options to delete, symlink, or stub those files.

## How It Works

### Scope

This skill is repo-specific.

- Only consider repo-local instruction files:
  - `AGENTS.md`
  - `CLAUDE.md` (repo root)
  - `.claude/CLAUDE.md`
  - `GEMINI.md` (repo root and/or repo subdirectories if present)
- Ignore system-wide instructions and files under the user's home directory (for example `~/.claude/`, `~/.gemini/`, `~/.config/`, etc.).

### Propose-first workflow (do not auto-edit)

1. Read `AGENTS.md`.
2. Identify the user's correction and translate it into a durable, repo-specific instruction.
3. Decide where it belongs:
   - Prefer editing the most relevant existing section.
   - If it introduces a genuinely new topic, propose a new `##` section near related content.
4. Draft the exact text change(s) as a small patch/snippet.
5. Ask for confirmation before applying:
   - "Apply these changes to `AGENTS.md`? (y/n)"

Only apply edits after the user confirms.

### Seamless integration rules

- Do not add dated changelog entries like "YYYY-MM-DD: correction".
- Update the existing prose/bullets/steps so the rule reads like it has always been part of the document.
- Keep formatting consistent with the surrounding `AGENTS.md` style.

### Contradiction checks

Before proposing a patch:

1. Scan `AGENTS.md` for conflicts with the new correction.
2. If a conflict exists, include conflict resolution in the proposed patch by rewriting/removing the conflicting lines.
3. If resolution is ambiguous, ask a single targeted question before proposing a patch.

Then scan repo-local guidance (if present):

- Compare `AGENTS.md` against `CLAUDE.md` and `.claude/CLAUDE.md`.
- Flag duplication and conflicts.
- Prefer consolidating into `AGENTS.md` as the canonical source of truth.

### Consolidation flow for CLAUDE.md files (if present)

If `CLAUDE.md` and/or `.claude/CLAUDE.md` exists in the repo:

1. Propose a consolidation patch to `AGENTS.md`.
2. After the user approves the `AGENTS.md` patch, propose what to do for each CLAUDE file:
   - Delete the file.
   - Replace the file with a symlink pointing to `AGENTS.md`.
   - Replace the file contents with a small stub that points humans/tools to `AGENTS.md` (fallback when symlinks are undesirable).
3. Confirm destructive actions (deletes) in a separate prompt.

Symlink directions (when chosen):

- `CLAUDE.md` -> `AGENTS.md`
- `.claude/CLAUDE.md` -> `../AGENTS.md`

### Notes for multi-instruction environments

Some harnesses merge multiple instruction sources and may apply conflicting rules non-deterministically. Prefer a single canonical source (`AGENTS.md`) and reduce parallel instruction files via consolidation + delete/symlink/stub.

Attribution

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

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.