Skip to content

AGENTS.md

Latest commit

 

History

History
36 lines (24 loc) · 1.91 KB

AGENTS.md

File metadata and controls

36 lines (24 loc) · 1.91 KB
Raw

AGENTS.md

Guidance for AI coding agents (and humans) working on this repository.

What this is

A Claude Code plugin: eight markdown skills that guide an external user from zero to a running, exposed app on the Intility Developer Platform. There is no application code — the "source" is SKILL.md prompts, YAML/JSON manifests, and reference docs.

Layout

.claude-plugin/plugin.json       plugin manifest (version is bumped by Release Please — don't edit by hand)
.claude-plugin/marketplace.json  lets this repo serve as its own marketplace
skills/<name>/SKILL.md           one skill per directory
skills/<name>/references/        templates and example conversations, loaded on demand

Conventions

  • Keep SKILL.md under 500 lines. Move templates, long examples, and walkthroughs to references/ and link them — they're loaded only when needed.
  • Frontmatter: description is written in third person and includes concrete trigger phrases. allowed-tools is the minimum the skill needs — prefer scoped patterns (Bash(oc delete httproute*)) over broad ones (Bash(oc delete*)); never pre-approve find, unscoped curl, or sed (use Glob/Read/Edit tools instead).
  • Skills detect state, they don't assume it. Every skill starts by querying (oc whoami, indev cluster list) rather than trusting conversation memory. Keep it that way.
  • Tone: plain language for Kubernetes beginners. One command at a time. Introduce a term after showing what it does.
  • Safety rails stay: internal gateway by default, two-step confirmation before public, status stays read-only, never create a second cluster without explicit double confirmation.

Before committing

claude plugin validate . --strict

This is also enforced by CI on every PR.

Commits

Conventional Commits (fix:, feat:, docs:, …) — Release Please derives versions from them. See CONTRIBUTING.md.