MaestroKit
0.10.0
MaestroKit - plan/implement/test/document workflow installer, validators and packaging for AI coding agents.
Minimum PowerShell version
7.0
Installation Options
Owners
Copyright
(c) 2026 Arthur Biazon Machado. MIT licensed.
Package Details
Author(s)
- Arthur Biazon Machado
Tags
AgentSkills AI ClaudeCode OpenCode OhMyPi Workflow
Functions
Install-Maestro Test-MaestroKit Invoke-MaestroSmokeInstall
PSEditions
Dependencies
This module has no dependencies.
Release Notes
## 0.10.0 — 2026-09-22 ### Added - **Plan intent layer**: Plans now carry an `Intent` section — Purpose, FR-numbered Outcomes (EARS grammar for conditional/error rules), and Non-Goals with rationale — with depth proportional to the Test Tier. Milestones tag the FRs they implement via `Covers:`; every FR must be covered by at least one milestone. EARS sentence patterns and input/output validation tables replace adjective-only prose in Business Logic. - **`cue` skill**: the named human approval gate. Runs the readiness audit (intent completeness, FR coverage, blocking questions, DAG integrity), presents a one-screen review digest, and flips `Approved: pending` → `approved` on the user's explicit sign-off. Draft is not approved; file existence is not approval. - **Plan approval gate (breaking)**: `orchestrate` refuses to execute any Plan whose `Approved` field does not read `approved` — a missing field counts as not approved. - **`critique` agent**: fresh-context reviewer spawned by orchestrate's new Phase-2 Critique Gate. Verifies FR coverage, `Why & Limits` boundaries, and Non-Goals compliance; classifies gaps as implementation (route to `tune`) or spec (route to the user with a successor-plan recommendation). Strictly read-only. - **Living current-state layer**: new working file `knowledge/system-behavior.md` (spec: `references/system-behavior.md`). `score` harvests shipped capabilities and contracts into it at every completed Plan — offered regardless of `Docs Affected`; `tune` updates it in place when a fix changes behavior. Executed Plans stay frozen as audit records; this file replaces them as living truth. - **Standing principles**: new working file `knowledge/principles.md` (spec: `references/principles.md`). `prelude` bootstraps it from confirmed ADRs and codebase conventions; `rehearse` and `tune` promote invariants as they crystallize; `compose` embeds the relevant rules in milestone `Must not` lines instead of restating them everywhere. - **Spec Gap classification**: Issue Prevention gains an optional gap class (`surfacing`/`capture`/`construction`/`verification`) so recurring root causes repair the workflow itself, not just the next Plan. `tune` writes it when the cause traces to the Plan. - **interlude signals**: the status report adds underspecification/regression flags — failed milestones with `Retries >= 3`, open Issues linked to Done plans, 🔒 unapproved drafts, and the ✅⏳ harvest backlog. ### Changed - Milestone headers gain a `Covers: [FR-###]` field (canonical form updated in every template and example); `play` status blocks now report `Requirements covered:`; `arrange` derives validation/error tests from the FR list and audits assertion strength. - `elaborate` completes the Intent layer before elaborating milestones, and resets `Approved` to `pending` when it modifies an already-approved Plan (re-approval required). - Plans Index entries carry `🔒` while unapproved (approval marker precedes docs markers); `cue` removes it on approval. - `prelude` now also bootstraps Principles (readiness report walks Phases 1–5); `rehearse` probes Purpose and crystallizes principles during the interview. - compose/orchestrate/score skill texts, checklists, and descriptions updated for the new contract; conventions.md gains the approval rules, 🔒 marker vocabulary, ownership-matrix rows for `cue`/`critique`, and the two new knowledge files. ### Migration - Run `/cue <plan-id>` once on every existing pending Plan before re-running `/orchestrate` — the approval gate fail-closes on Plans without `Approved: approved`. - Existing executed Plans keep working as audit records; no Intent/Covers backfill is required for them.
FileList
- MaestroKit.nuspec
- tools\install-from-web.ps1
- tools\smoke-install.ps1
- CHANGELOG.md
- MaestroKit.psd1
- agents\critique.md
- references\contexts.md
- references\issues-index.md
- references\references-map.md
- references\testing-principles.md
- skills\compose\SKILL.md
- skills\interlude\SKILL.md
- skills\score\SKILL.md
- tools\publish-gallery.ps1
- tools\sync-supporters.ps1
- Install-Maestro.ps1
- MaestroKit.psm1
- agents\play.md
- references\conventions.md
- references\plan.md
- references\repo-fingerprint.md
- references\testing-tech-preferences.md
- skills\cue\SKILL.md
- skills\orchestrate\SKILL.md
- tools\add-supporter.ps1
- tools\sign-release.ps1
- tools\validate-bundle.ps1
- LICENSE
- README.md
- agents\tune.md
- references\design-principles.md
- references\plans-index.md
- references\system-behavior.md
- skills\arrange\SKILL.md
- skills\elaborate\SKILL.md
- skills\prelude\SKILL.md
- tools\build-module.ps1
- VERSION
- references\adrs.md
- references\issue.md
- references\principles.md
- references\tech-preferences.md
- skills\audition\SKILL.md
- skills\instruments\SKILL.md
- skills\rehearse\SKILL.md