Memory, code graph, messaging, files, canvases, secrets, safe SQL, support cases and SEO audits. One plugin gives a Claude Code agent all of them; the format is open, so other agents can read the same SKILL.md files.
A skill is a folder with a SKILL.md file in the open Agent Skills Standard. Its description says what the skill does and when to use it, so the agent loads it when the task calls for it and ignores it otherwise.
Markdown plus small scripts. No runtime to run and no server to host.
Only the short description sits in the agent's context. The body loads when the description matches the task.
The plugin is assembled by the plugin builder from local sources and from separate repositories, so each skill is versioned on its own.
Grouped by what they are for. Each card names what the skill needs and where its source lives.
Search what the agent learned in past sessions: decisions, facts, corrections, and what depends on an entity (--about).
Needs: AI Maestro running
Source: ai-maestro-plugins (src)
Look up a function, class or API in the generated documentation before calling it: signature, parameters, return type.
Needs: AI Maestro running
Source: ai-maestro-plugins (src)
Query the code graph for callers, dependencies and the blast radius of a change.
Needs: AI Maestro running
Source: ai-maestro-plugins (src)
Send, read and reply to signed messages between agents with the Agent Messaging Protocol. Local and federated delivery, attachments, Ed25519 signatures.
Needs: curl, jq, openssl; AI Maestro for local delivery
Source: agentmessaging/claude-plugin
Authenticate an agent with an auth server using Agent Identity: Ed25519 identity documents, proof of possession, OAuth 2.0 token exchange.
Needs: curl, jq, openssl 3.x
Source: agentmessaging/agent-identity
Share a file between agents, hosts and people through an AFP space (S3-compatible storage). Store, fetch, list and make time-limited links.
Needs: curl, jq, openssl; a configured AFP space
Source: agentmessaging/agent-files
Create interactive HTML pages (canvases) the person sees in the dashboard, and handle what they click, submit or select.
Needs: AI Maestro dashboard
Source: agentmessaging/agent-actions
Create, list, rename, hibernate, wake, export and delete agents, and install plugins on them, with the aimaestro-agent.sh CLI.
Needs: AI Maestro running
Source: ai-maestro-plugins (src)
Use an API key, token or password without ever seeing it, and ask the person to store one without pasting it into the chat. Backed by the aim-secret CLI and a local vault.
Needs: aim-secret (AI Maestro installs it)
Source: ai-maestro-plugins (src)
Run SQL against a production or staging database safely: check the saved query library first, run only read-only queries through the safe-psql wrapper (read-only role, SQL check, LIMIT added), save findings back to the library. Changing data is a separate door, safe-psql-write: it proposes one change, shows a dry run that is rolled back, and runs it only after a person approves that exact statement in a terminal.
Needs: safe-psql and safe-psql-write (AI Maestro sets them up), a read-only Postgres role, and for writes a settings file and a separate limited write login
Manage customer support cases with a written trail: a numbered case per report, what was found, how it was resolved, and runbooks for problems that come back.
Needs: Nothing beyond the agent
Audit and implement SEO for a website: meta tags, canonical URLs, Open Graph, X/Twitter cards, JSON-LD, sitemap and robots controls. Includes a build audit script.
Needs: Python 3 for the audit script
Persistent task_plan.md, findings.md and progress.md files so an agent stays on course through a long, multi-step task.
Needs: Bash only; works standalone
Source: ai-maestro-plugins (src)
With aim-secret, the value lives in a local vault (macOS Keychain, Linux libsecret, or an encrypted file). The agent works by name.
In Claude Code, the ai-maestro-secrets mod does the same with a form in the terminal. Scrubbing stops accidents, not an agent that wants the value, so use narrowly scoped keys. Details in SECRETS.md and CLAUDE-CODE-MODS.md.
The AI Maestro installer and updater install the plugin for you. You can also add it by itself.
The installer sets up the dashboard, the messaging scripts, the plugin and the helpers the skills need (aim-secret, safe-psql).
Skills marked "AI Maestro running" need the AI Maestro service on localhost:23000. planning and agentic-support work without it.
The skills follow the Agent Skills Standard, so any agent that reads that format can use a SKILL.md folder copied from the plugin repository. The scripts that talk to AI Maestro need the service either way.
Thirteen in plugin 1.8.0: agent-files, agent-identity, agent-messaging, agentic-seo, agentic-sql, agentic-support, ai-maestro-agents-management, aim-secret-management, canvas-actions, docs-search, graph-query, memory-search and planning.
Yes. The plugin builder assembles the plugin from a manifest, so you can drop skills, add your own, or pull one from another repository.
Claude Code, Codex and Grok Build have first-class support. Any other terminal agent runs in the terminal tab. The plugin install above is for Claude Code.
Install AI Maestro, or read how the open protocols behind the messaging, identity, actions and files skills fit together.