Changelog

v0.3.0

August 22, 2026

Canonical agent directives are ~24% leaner at the source (~21% as shipped in generated plugins): the bounded-repair contract lives once in the global rules and specialist skill catalogs list only verified skills.
Delegated implementation outcomes now complete at a pushed feature branch with an open PR; routine commit, push, and PR creation require no approval asks, while merge, release, and production actions remain separate authorization boundaries.
Failed or cancelled delegations get one adjusted-brief retry before structured blocker reporting.
Decision record: ADR-CORE-019 (docs/adr/core/ADR-CORE-019-directive-simplification.md).

The published package can be installed, checked, updated, and removed with maestria install claude-code, maestria status, maestria update claude-code, and maestria uninstall claude-code.
The Maestria CLI stages the npm package in a local marketplace and delegates installation state to Claude Code’s native plugin manager.
Exact version pinning is not supported for maestria update claude-code --version; the host marketplace selects the latest package.
Plugin metadata now stays aligned with package versions: release-time version sync is generalized across plugins and a --check mode fails the build on drift, so the published plugin artifact can no longer diverge from npm metadata.
Agent directives now calibrate effort to task risk, prefer mature ecosystem solutions, converge reviews on material blockers, deliver routine engineering work through feature branches and PRs, and clean up task-owned background processes before completion.

Initial release

@maestria/claude-code is a declarative Claude Code plugin generated from the canonical directives in packages/core/agent-directives/:

7 specialist agents with global-rules skill preloading - adventurer, architect, builder, diagnose, planner, reviewer, writer - available as @maestria:<agent> mentions
Orchestrator and global-rules skills - /maestria:orchestrator routes work to specialists; maestria:global-rules is preloaded into every agent
Workflow commands - /maestria:fein, /maestria:sonar, and /maestria:blitz select the pipeline depth
Read-only enforcement - adventurer, planner, and reviewer deny Write/Edit via disallowedTools (user-authorized; the only runtime enforcement)
Local validation - claude plugin validate --strict is the automated gate; marketplace installation is available through Claude Code or the Maestria CLI

The plugin tracks the shared agent directives as they evolve. This release reflects the latest canonical contracts:

Clarified agent workflow contracts while preserving detailed specialist guidance: routine validated commits on recognized feature branches remain autonomous after required review; push and later lifecycle actions stay separately gated. Added bounded repair and platform-enforcement notes, refreshed generated projections, and retained explicit mode reset behavior and read-only sonar profiles where supported.
Simplified and re-aligned the shared agent directives around outcome, evidence, runtime authority, blind review, bounded repair, and autonomous routine work. Directives now avoid unnecessary orchestration ceremony, allow the host runtime to determine whether work is performed directly or delegated, prevent nested supervisors from duplicating scheduling and lifecycle work, and keep all generated platform projections synchronized.