Skip to content
Back to skills

Myco Rules

ASecurity

Keep AGENTS.md minimal, durable and canonical across every agent that reads it. Adds, trims and audits project rules without letting the rules file decay into a dump of whatever was true last week.

  • 14 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 19, 2026
ai-agentsgogit

Security analysis

A100/100

Pro scans all 3 files and shows the line behind each finding

Scanned September 19, 2026

npx -y skills add goondocks-co/myco --skill myco-rules --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Myco Rules?

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

Security grade badge for Myco Rules
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/goondocks-co-myco-rules/badge)](https://www.skillsdirectory.com/skills/goondocks-co-myco-rules)

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: myco-rules
description: >-
  Keep AGENTS.md minimal, durable and canonical across every agent that reads
  it. Adds, trims and audits project rules without letting the rules file decay
  into a dump of whatever was true last week.
when_to_use: >-
  "add a project rule", "trim AGENTS.md", "audit our rules", or whenever a rule
  is about to be written down and it is not obvious whether it belongs in
  AGENTS.md or in Myco as a spore.
allowed-tools: Read, Edit, Write, Grep, Glob
user-invocable: true
---

# Myco Rules — Minimal Durable Rules

Use this skill to keep `AGENTS.md` sharp, durable, and lightweight.

Do not duplicate changing project context in static rules files. Keep the rules surface small and stable.

## Goals

- Keep `AGENTS.md` as the canonical rules file
- Keep agent-specific files as thin stubs that point back to `AGENTS.md`
- Add only rules that are durable enough to survive normal project change
- Remove vague, stale, duplicated, or over-detailed guidance

## What Belongs in `AGENTS.md`

- Project identity and scope
- Non-goals and hard boundaries
- A small number of architecture invariants
- Quality gates and required verification commands
- Brief working-style guidance that is stable across branches

## What Does Not Belong

- Detailed golden paths that drift with the codebase
- Current initiative notes, migration steps, or branch-specific instructions
- Long architecture tours or other context better delivered dynamically
- Decision rationale or recent history that belongs in project memory, not the rules file
- Workflow history or state better preserved by version control, issue trackers, or project memory
- Generic filler such as "follow best practices" or "keep things clean"

## Placement Rules

- Default to editing `AGENTS.md`
- Treat `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md` as thin stubs unless the developer explicitly wants agent-specific behavior there
- If a project has important generic rules in `CLAUDE.md` but not `AGENTS.md`, suggest migrating them into `AGENTS.md`

## Audit Checklist

When reviewing a rules file, look for:

- Vague language: "best practices", "when possible", "try to", "keep it clean"
- Stale anchors: files, commands, or paths that no longer exist
- Bloat: repeated guidance better delivered as dynamic context
- Duplication: the same rule spread across `AGENTS.md` and agent-specific stubs
- Missing hard boundaries: no non-goals, no quality gate, no canonical source of truth

Prefer deleting stale or weak rules over replacing them with more prose.

## Choosing Rule Candidates

Promote only guidance that deserves to live in `AGENTS.md` for a long time.

Good candidates:

- Repeated gotchas with the same root cause
- Repeated bug fixes caused by violating an unwritten invariant
- Explicit decisions that imply a stable always/never rule

Do not promote:

- One-off issues
- Recent feature work that may still change
- Context that is already better delivered dynamically during the session

## Editing Workflow

1. Discover `AGENTS.md` and any agent-specific instruction files.
2. Treat `AGENTS.md` as canonical by default.
3. Identify rules to remove, simplify, or add.
4. Propose the minimal diff that improves clarity.
5. Apply approved edits.

## Rule Writing Standard

- Be specific enough to violate
- Prefer direct language: `MUST`, `MUST NOT`, `SHOULD`
- Anchor claims to real files or commands when needed
- Keep the file short enough that an agent will actually read it
- Treat code comments with the same discipline: keep them brief, durable, and focused on helping someone understand non-obvious code

## References

- `references/rules-good-example.md`
- `references/rules-bad-example.md`

Files in this skill

  • SKILL.md3.6 KB
  • references/rules-bad-example.md5.4 KB
  • references/rules-good-example.md3.5 KB

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.