PM AI Toolkit

A plugin for Cursor and Claude Code that turns raw PM inputs — FSD, BRD, PRD documents, Figma designs, images, OpenAPI specs, and email files (.eml) — into a full suite of development artifacts: requirements, features, user stories, tasks, test cases, exit criteria, architecture diagrams, sprint plans, and stakeholder updates. Built-in version tracking, scope analysis, delta detection, and rollback.

Quick Start

# Install
npx pm-ai-toolkit install

# Restart Cursor, then kick off a new project
Use pm-ai-toolkit:brainstorming
# — or, if you already have documents —
Use pm-ai-toolkit:knowledge-ingestion

What's Included

Skills (16)

| Skill | Triggers When | What It Does | |-------|---------------|--------------| | pipeline-orchestrator | PM wants to run the full pipeline end-to-end | Runs all skills in dependency order with configurable approval gates, progress tracking, and resume support via .pipeline-state.json | | status-dashboard | PM asks "what's the status?" or returns to a project | Read-only scan of all artifacts: versions, pipeline completeness, staleness warnings, and recommended next actions | | brainstorming | Starting a new initiative or exploring ideas | Guided ideation: explores approaches, clarifies requirements, produces an approved design spec, then hands off to knowledge-ingestion | | knowledge-ingestion | PM provides FSD, BRD, PRD, Figma links, images, API specs, or email files (.eml) | Parses diverse inputs (including stakeholder email threads) into a normalized requirements summary with gap analysis and RFC 2119 language normalization | | feature-extraction | After requirements are ingested | Transforms requirements into a prioritized feature list with MoSCoW classification, dependencies, and tech stack detection | | user-story-generation | After features are extracted | Converts features into user stories with personas, story points, acceptance criteria, and Definition of Ready checklist | | task-decomposition | After user stories are generated | Breaks stories into granular 2–4 hour development tasks with effort estimates and layer assignments | | test-case-generation | After tasks are decomposed | Generates unit/integration/e2e test cases with code snippets, coverage matrix, exit criteria cross-references, and optional Excel export | | exit-criteria-generation | After tasks are decomposed | Produces BDD acceptance scenarios and verification checklists per feature, cross-linked to test case IDs | | architecture-diagrams | After features are extracted | Generates Graphviz DOT or Mermaid diagrams: solution overview, technical architecture, class, sequence, and ER diagrams — with syntax validation | | scope-analyzer | Before any regeneration | Delta analysis engine: detects changes in inputs, traces downstream impact, flags possibly-removed requirements, and produces a scope delta report | | risk-register | After agent reviews or when PM raises risks | Aggregates project risks from all artifacts, scores by likelihood × impact, defines mitigations, and tracks risk lifecycle | | version-changelog | After any skill generates artifacts | Manages semantic versioning, draft/finalized lifecycle, archiving (<artifact>-v<version>.md), and master CHANGELOG.md | | sprint-planner | After user stories are generated, or when PM says "plan sprints" | Allocates stories to numbered sprints by velocity, MoSCoW priority, and capacity — with utilisation warnings and optional release milestones | | stakeholder-update | When PM needs to communicate progress | Generates status updates in exec-brief, engineering-detail, or customer-facing tone from current artifact state | | rollback | When PM needs to restore a previous finalized version | Lists archived versions, restores selected version as new draft baseline, logs to CHANGELOG, and runs scope analysis |

Agents (5)

Agents are dispatched at key pipeline gates — always optional, never forced. Each agent provides a structured review report that feeds back into the pipeline severity system (BLOCKER / WARNING / INFO).

| Agent | Dispatched After | What It Reviews | |-------|-----------------|-----------------| | spec-reviewer | Brainstorming produces a design spec | Completeness, consistency, scope, and pipeline-readiness of the spec | | requirements-analyst | Stage 1: knowledge-ingestion | Requirements for completeness, consistency, ambiguity, testability, and feasibility | | story-reviewer | Stage 3a: user-story-generation | User story quality against INVEST criteria, acceptance criteria testability, and persona alignment | | solution-architect | Stage 3b: architecture-diagrams | Architecture for soundness, scalability, security, separation of concerns, and technology fit | | task-estimator | Stage 4: task-decomposition | Effort estimates for uncertainty (Three-Point Estimation), splitting candidates, and schedule risk |

Pipeline Flow

[Entry Points]
  brainstorming ──────────────────────────────────────────────────────────┐
  knowledge-ingestion (direct entry if documents/emails are ready)          │
                                                                           │
[Stage 1] knowledge-ingestion                                             ←┘
            │  optional: spec-reviewer agent (before stage 1)
            │  optional: requirements-analyst agent (after stage 1)
            ▼
[Stage 2] feature-extraction
            ├──────────────────────────────────────────────────────────────┐
            ▼                                                               ▼
[Stage 3a] user-story-generation              [Stage 3b] architecture-diagrams
            │  optional: story-reviewer agent              │  optional: solution-architect agent
            ▼                                               ▼
[Stage 4] task-decomposition
            │  optional: task-estimator agent
            ├─────────────────────────────────────────────────────────────┐
            ▼                                                              ▼
[Stage 5a] test-case-generation         [Stage 5b] exit-criteria-generation
  (includes exit_criteria_ref)            (includes test_coverage column)
            │                                              │
            └──────────────────────┬───────────────────────┘
                                   ▼
[Stage 6] version-changelog (draft tracking after every stage)

[Optional]
  sprint-planner    ← plan sprints after user stories exist
  stakeholder-update ← generate comms at any time
  rollback          ← restore previous finalized version

[Cross-cutting — invoked automatically]
  scope-analyzer    ← runs before any regeneration to detect changes
  risk-register     ← aggregates risks from all artifacts and agent reviews
  pipeline-orchestrator ← meta-skill driving stages 1–6 end-to-end

Version Lifecycle

All artifacts follow a three-state lifecycle controlled by the PM:

1. DRAFT — Artifacts are generated and freely iterated. Edits overwrite in place.
2. FINALIZED — PM says "finalize this version". Artifacts archived to docs/pm-artifacts/archive/<artifact>-v<version>.md, CHANGELOG updated, version locked.
3. NEW ITERATION — PM provides updated inputs. scope-analyzer runs delta analysis against the finalized baseline before any regeneration.

Installation

Via npm (recommended)

npx pm-ai-toolkit install

Or install the CLI globally:

npm install -g pm-ai-toolkit
pm-ai-toolkit install

This copies all plugin files (skills, agents, hooks) into ~/.cursor/plugins/local/pm-ai-toolkit/. Restart Cursor or Claude Code to activate.

CLI Commands

| Command | Description | |---------|-------------| | pm-ai-toolkit install | Install or update the plugin | | pm-ai-toolkit update | Alias for install | | pm-ai-toolkit uninstall | Remove the plugin | | pm-ai-toolkit status | Show installed version and check for updates | | pm-ai-toolkit doctor | Run diagnostics: Node version, Cursor dir, file integrity, skill/agent counts | | pm-ai-toolkit init | Create a pm-toolkit.config.yaml in the current project with commented defaults | | pm-ai-toolkit validate-config | Validate pm-toolkit.config.yaml — exits with code 1 if errors found | | pm-ai-toolkit project-status | Show all artifact files, versions, and statuses for the current project (--json for machine-readable output) | | pm-ai-toolkit help | Print usage information | | pm-ai-toolkit --version | Print version number |

Platform Support

| Platform | npm CLI | install.sh | |----------|---------|------------| | macOS | ✓ | ✓ | | Linux | ✓ | ✓ | | Windows (native) | ✓ | ✗ (use npm) | | Windows (WSL) | ✓ | ✓ |

Via script

git clone <repo-url> && cd pm-ai-toolkit && ./install.sh

Configuration

Copy the example config to your project root to customize behavior:

pm-ai-toolkit init

Key settings in pm-toolkit.config.yaml:

output_dir: docs/pm-artifacts/     # Where artifacts are saved
diagram_format: both                # dot | mermaid | both
excel_export: auto                  # auto | always | never
tech_stack: ""                      # Pre-set stack; skips auto-detection
project_name: ""                    # Used in artifact filenames
stakeholder_update_tone: ask        # exec | engineering | customer | ask
auto_approve_stages: []             # Skills that skip PM approval gate

The config file is validated by a JSON Schema (pm-toolkit.config.schema.json) — editors with YAML Language Server support will provide autocomplete and inline validation automatically.

Explicit Invocation

Skills activate automatically based on context. You can also invoke them directly:

Use pm-ai-toolkit:pipeline-orchestrator
Use pm-ai-toolkit:status-dashboard
Use pm-ai-toolkit:brainstorming
Use pm-ai-toolkit:knowledge-ingestion
Use pm-ai-toolkit:feature-extraction
Use pm-ai-toolkit:user-story-generation
Use pm-ai-toolkit:task-decomposition
Use pm-ai-toolkit:test-case-generation
Use pm-ai-toolkit:exit-criteria-generation
Use pm-ai-toolkit:architecture-diagrams
Use pm-ai-toolkit:scope-analyzer
Use pm-ai-toolkit:risk-register
Use pm-ai-toolkit:version-changelog
Use pm-ai-toolkit:sprint-planner
Use pm-ai-toolkit:stakeholder-update
Use pm-ai-toolkit:rollback

Directory Structure

pm-ai-toolkit/
  package.json
  pm-toolkit.config.example.yaml
  pm-toolkit.config.schema.json
  bin/
    cli.js
  .cursor-plugin/
    plugin.json
  .claude-plugin/
    plugin.json
    marketplace.json
  hooks/
    hooks.json
    hooks-cursor.json
    run-hook.cmd
    session-start
  skills/
    _shared/
      conventions.md
    pipeline-orchestrator/
      SKILL.md
    status-dashboard/
      SKILL.md
    brainstorming/
      SKILL.md
    knowledge-ingestion/
      SKILL.md
    feature-extraction/
      SKILL.md
    user-story-generation/
      SKILL.md
      story-templates.md
    task-decomposition/
      SKILL.md
    test-case-generation/
      SKILL.md
      test-patterns.md
      excel-export.md
    exit-criteria-generation/
      SKILL.md
      criteria-templates.md
    architecture-diagrams/
      SKILL.md
      diagram-patterns.dot
      diagram-templates.md
      mermaid-templates.md
    scope-analyzer/
      SKILL.md
    risk-register/
      SKILL.md
    version-changelog/
      SKILL.md
      changelog-format.md
    sprint-planner/
      SKILL.md
    stakeholder-update/
      SKILL.md
    rollback/
      SKILL.md
  agents/
    requirements-analyst.md
    solution-architect.md
    spec-reviewer.md
    story-reviewer.md
    task-estimator.md
  tests/
    cli.test.js
    skills.test.js
  README.md
  CHANGELOG.md
  HOOKS.md
  INSTALLATION.md
  GETTING-STARTED.md
  CONTRIBUTING.md
  ARTIFACT-SCHEMA.md
  PLUGIN-REVIEW.md

License

MIT