Focus Consulting's private Claude Code plugin marketplace. Operationalizes Focus's AI-Native Operating Model into a set of installable plugins for engagement teams.
Audience: Focus consultants and engineers on Focus engagement teams. This marketplace is private to the focusconsulting GitHub org.
Five plugins, each scoped to a role or concern:
| Plugin | Surface | Owner roles |
|---|---|---|
| focus-corpus | Corpus mechanics: docling-based ingestion (bundled MCP server + local-CLI fallback), evidence search, stale-artifact monitor, the /focus-pages doc renderer |
Infrastructure (everyone) |
| focus-pm | User-story authoring, outcomes stewardship, sprint planning, stakeholder updates | PM |
| focus-designer | Spec authoring, evidence inventory, comprehension review, five research-archetype agents | Designer |
| focus-engineer | Plan + ADR authoring, contract tests, validate-plan, security + pre-merge hooks, codebase agents | Lead Engineer + Product Engineer |
| focus-delivery | Client-facing sprint reports | Engagement Lead + PM |
A complete inventory: 15 skills, 20 commands, 8 agents, 7 hooks + 1 monitor, 9 templates, ~38 doc pages.
See INSTALL.md for the engagement-team install instructions. The recommended path is per-engagement-repo via .claude/settings.json so the whole team gets the same plugin set automatically.
Focus's AI-Native Operating Model defines:
- Four roles — Product Manager, Designer, Lead Engineer (System Owner), Product Engineer
- Sprints — 2 weeks each, back-to-back. Discovery happens before sprints; sprints run continuously thereafter (no between-sprint cool-down period).
- Six failure modes the encoded-intent layer fights — plausibility collapse, coherence debt, verification debt, production-outpacing-judgment, review-capacity crisis, framing capture, spec/agent drift, comprehension drift
- Asymmetric AI roles per artifact — reference-only for constitutions, lookup-only during sprint planning, draft+tradeoff for ADRs, primary-generation for code
The marketplace uses Agile vocabulary (user story, points, sprint, sprint planning, progress reporting). The operating model underneath is shaped differently from Agile defaults — see VOCABULARY.md for what each term means here and what behavioral expectations from "real Agile" do not apply.
Every plugin in this marketplace exists to make one of those moves easier and more reliable. The asymmetry between roles is intentional: a PM cannot auto-generate user stories via these plugins; a Designer cannot auto-generate spec acceptance criteria. The encoded-intent layer is the asymmetry.
focus-claude-plugins/
├── .claude-plugin/marketplace.json # Marketplace manifest (5 plugins)
├── focus-corpus/ # Per-plugin directory
│ ├── .claude-plugin/plugin.json
│ ├── skills/<name>/SKILL.md # Skills (LLM-invokable patterns)
│ ├── commands/<name>.md # Slash commands
│ ├── agents/<name>.md # Sub-agents
│ ├── hooks/ # Hook scripts + hooks.json
│ ├── templates/ # Author-fillable templates
│ ├── docs/{guides,essays}/ # Source markdown
│ ├── docs/index.html # Rendered single-page doc site
│ ├── scripts/ # Plugin-internal scripts (corpus only)
│ └── mcp-server/ # Bundled docling MCP server (corpus only; Python)
├── focus-pm/, focus-designer/, focus-engineer/, focus-delivery/
└── focus-discovery/, focus-brownfield/, focus-bd/ # Staged for v2
The three v2-staged plugin directories exist but are not advertised in the marketplace yet. They'll be enabled once their content lands.
Every plugin's docs/index.html is a self-contained single-page site rendered by focus-corpus/scripts/render-pages.mjs. Open it in any browser. The site reads guides + essays + plugin manifest from disk; no build step beyond node focus-corpus/scripts/render-pages.mjs <plugin-dir>.
A few load-bearing principles that shaped every plugin in this marketplace. Worth knowing before you author a new skill or modify an existing one:
- Output discipline. Every skill prompt declares: no preamble, no recap, no summary tail. Every skill ends with a
## Failure modes this defends againstblock citing V0 contract failures. - Read-only by design. Verification skills never modify artifacts. Author-side commands scaffold only — substance is human authorship per the operating model's authorship boundaries.
- Default to green. RAG-style audits prefer GREEN under ambiguity. False positives erode trust faster than false negatives reveal drift.
- Hardcoded operating-model paths. No corpus-root configuration; no ADR-directory configuration; no plan-directory configuration. The operating model has one canonical layout.
- Hook-vs-skill split. Hooks do the fast critical-path subset of each skill's audit. Skills are the comprehensive surface.
- Source attribution. Every lifted asset (arckit hooks, checkin commands, Anthropic skills, research-corpus-mcp agents) carries a header citing source + license + date.
- Single-shot sparring stance. Authoring commands offer ONE optional question pair before scaffolding. Multi-turn co-drafting risks framing capture.
Private to Focus Consulting. Unlicensed for external distribution.
Vendored components carry their original licenses + attribution headers:
- arckit security hooks + monitor pattern: MIT, https://github.com/tractorjuice/arc-kit
- knowledge-work-plugins (Anthropic): Apache 2.0
- research-corpus-mcp: internal Focus tool (archived — superseded by
focus-corpus/mcp-server/) - checkin commands: internal Focus tool
0.2.0. Vocabulary rewrite from Shape Up to Agile (see VOCABULARY.md). The operating model is unchanged. Two sources of breaking change for engagements upgrading from 0.1.0: (a) hardcoded paths moved from product/pitches/ to product/stories/, and (b) frontmatter field names changed (appetite → points, pitch: → story:, etc.). Engagements with existing 0.1.0 artifacts must migrate manually or pin 0.1.0.