Maestria Skills for Prime Agent

@maestria/prime-agent encodes the Maestria engineering methodology for Prime Agent: 7 specialist roles, an orchestrator, the global-rules contract, handoff and iteration-limits aids, and the fein/sonar/blitz workflow modes - delivered as standard Agent Skills (skills/<name>/SKILL.md), generated from the canonical directives in packages/core/agent-directives/ - plus a small, verified Prime/Pi extension (dist/extension.mjs) for workflow-mode commands and mode prompt injection.

Native candidate - verified subset only

The generated skills match the documented Prime Agent Agent Skills contract and the extension is verified against the pinned Prime fork’s public extension API, but runtime behavior is not yet tested end to end in a live Prime session. Native recursive-subagent (rlm) dispatch and JSON/RPC headless-mode integration remain deferred. This package is not a production support promise and does not claim runtime enforcement.

Install

The preferred path is to register the published npm package with Prime:

prime-agent package install npm:@maestria/prime-agent

For skills-only installs, point Prime at the package’s skills/ directory in settings, or copy or symlink the skill directories into a project/global skill location. See the installation guide for the full options (including how to enable the extension for non-package installs).

Installation & SetupRegister the package, enable the extension, and verify the skills.

What It Provides

• 14 Agent Skills (skills/<name>/SKILL.md) with the required name/description frontmatter:
  • 7 specialists: adventurer, architect, builder, diagnose, planner, reviewer, writer
  • orchestration/rules: orchestrator, global-rules, handoff, iteration-limits
  • workflow modes: fein, sonar, blitz
• A verified Prime/Pi extension (dist/extension.mjs) registering /fein, /sonar, /blitz, /mode-clear, and /maestria-status, and injecting the active mode’s prompt on every agent turn.

Platform notes and limitations

• Verified subset only, not native rlm dispatch: the extension covers mode commands and mode prompt injection. Prime’s recursive-subagent (rlm(...)) call is an IPython-side tool with no public JS extension bridge, so this package cannot spawn child agents from the extension.
• Advisory, not enforced: skills, rules, and the extension are advisory guidance, not security enforcement. Prime Agent has no skill-level tool-denial mechanism, so read-only roles state their role intent without claiming a runtime boundary.
• Not a sandbox: Prime Agent executes model-generated Python and project commands with your user permissions. Restrict use to trusted repositories, skills, and instructions.
• No filesystem writes: the extension writes nothing (no ~/.pi, no .prime/agent writes); mode state rides on host session entries.

Explore the Docs

Getting StartedRegister the package with Prime, enable the extension, and verify skill discovery.

Quick StartLoad the orchestrator and specialist skills on a real Prime Agent task.

ChangelogRelease history for @maestria/prime-agent.

ContributingUnderstand the generated Agent Skills projection and the sync pipeline.