Featured article
Article
An agent is a named role inside a setup
An expanded guide to agent: specialisation, inputs, tools, limits, and a verifiable result.
Read article →Articles
Articles on building safer coding-agent setups, plus an exact record of what changed.
Article types
- Article
- Field note
- Changelog
- Release notes
Latest articles
Article
A command is a named, verifiable operation
An expanded guide to command: invocation interface, parameters, repeatability, and responsibility boundaries.
Article
A hook is an automatic action on a lifecycle event
An expanded guide to hook: events, argv, determinism, side effects, and rollback.
Article
Standing rules are a component, not an agent role
An expanded guide to instruction: boundaries, precedence, native files, and safe distribution.
Article
MCP is an external tool interface, not a secret config
An expanded guide to MCP: server, client configuration, transport, permissions, and boundary checks.
Article
A plugin is a package that extends the harness
An expanded guide to plugin: manifest, package contents, harness projections, and installation trust.
Article
A setting is non-secret behavioural configuration
An expanded guide to setting: config files, scope, MCP names, and the no-secrets rule.
Article
A skill is a workflow package, not a prompt in a folder
An expanded guide to skill: triggers, procedures, progressive disclosure, and the supply-chain surface.
Article
Antigravity CLI
A Gemini-based harness with skills, agents, plugins, hooks, MCP, and project resources
Article
Claude Code
A terminal coding agent with CLAUDE.md, skills, agents, hooks, MCP, and versioned plugins
Article
Codex
A coding agent with AGENTS.md, config.toml, subagents, hooks, and plugin projection
Article
Cursor
An IDE harness with .cursor-plugin/plugin.json, rules, skills, agents, hooks, MCP, and commands
Article
Grok Build
An xAI harness with AGENTS.md, skills, plugins, hooks, MCP, and config.toml
Article
MiMoCode
A system from the NDDev OpenNetwork line; ai-stp does not currently claim it as a separate supported harness.
Article
OpenCode
An open harness with native skills, plugins, agents, commands, MCP, and JSON configuration
Article
Pi
A minimal harness with package resources, skills, extensions, prompts, and target settings
Field note
The AI-native operating model and the project function
Why AI-native work starts with process, context, and ownership rather than model choice.
Field note
An epitaph for old-school vibe coders
Why prototyping without system ownership becomes an expensive theatre of misunderstanding.
Field note
How to write Agent Skills without making an agent cosplay a junior
A practical field note on skills as workflows, not folders of Markdown.
Field note
Vibe-coding safety: keep an agent from leaking data to production
A practical review of secret leaks, protection layers, and safe work with AI agents.