Releases: Maximepodgorski/agent-skills
Release list
v1.4.0 — Component Skill: Ship Action + Writing-Plans Spec
Component skill upgraded to v1.2: the dev action becomes ship, the spec doubles as an executable plan, and the new ship loop runs task-by-task in TDD with a choice between subagent-driven and inline execution.
New Files
| File | Description |
|---|---|
component/references/actions/ship.md |
Ship action — task-by-task TDD execution loop with PLAN / ONE-SHOT modes, subagent vs inline mode choice, spec drift handling, stuck detection |
component/references/templates/ship-output.md |
Chat output format for ship — per-task block with checked steps, checkpoint format, 5 exit states (Complete / Partial / Stuck / Blocked / Spec Missing Tasks) |
Removed Files
| File | Reason |
|---|---|
component/references/actions/dev.md |
Renamed to ship.md |
component/references/templates/dev-output.md |
Renamed to ship-output.md |
Changed Files
| File | Change |
|---|---|
component/SKILL.md |
Renamed dev → ship across action table, action router, output templates, references; bumped version to 1.2 |
component/CHANGELOG.md |
Added v1.2 entry — rename, writing-plans pattern, subagent/inline modes, migration notes |
component/README.md |
Updated quick start, action table, and recommended workflow for ship |
component/references/actions/spec.md |
Added Step 4 (generate Implementation Tasks) + Step 5 (self-review checklist); updated spec generation order to include Header, Token Mapping, Blockers, Tasks |
component/references/templates/spec.md |
Added header (Goal / Architecture / Tech Stack), Create / Modify / Test File Structure, Implementation Tasks section with TDD bite-sized steps, no-placeholders rule, self-review checklist |
component/references/actions/review.md |
Updated next-step pointer dev → ship |
component/references/actions/spec-review.md |
Updated review focus pointer dev → ship |
component/references/templates/doc-output.md |
Updated next-step pointer to ship |
component/references/templates/review-output.md |
Updated next-step pointer to ship |
component/references/templates/spec-output.md |
Updated next-step pointer to ship |
component/references/templates/spec-review-output.md |
Updated next-step pointers to ship |
CLAUDE.md |
Updated component skill action list and project structure |
README.md |
Updated component skill quick start and action table |
Summary
- Renamed component action
dev→shipacross all files, cross-refs, and external docs - Added Implementation Tasks section to spec template — bite-sized TDD steps with exact files, code, commands, commits (writing-plans / executing-plans pattern from superpowers)
- Added spec Header (Goal / Architecture / Tech Stack) and
Create / Modify / TestFile Structure - Added spec self-review checklist (coverage, placeholders, type consistency, tokens, command accuracy, granularity)
- Rewrote ship action as a task-by-task TDD loop consuming the spec's Implementation Tasks section
- Added execution mode choice to ship: Subagent-Driven (recommended >5 tasks) via
superpowers:subagent-driven-development, or Inline (recommended ≤5 tasks) with checkpoints every 2-3 tasks - Added spec drift handling — deviations logged in spec Notes, structural drift surfaces to user
- Added resumability — checkbox state persists to spec, interrupted sessions resume at first unchecked step
- Added new Spec Missing Tasks exit state for legacy v1.1 specs without Implementation Tasks
v1.3.0 — Design Screen Skill + Craft Action
New skill: design-screen — compose UI screens from existing components with ASCII wireframe proposals, component coverage tracking, and a Figma placement draft step. Includes 6 actions (init, spec, spec-review, craft, ship, review), 4 principles, and 8 templates. Also added Deep Context section to kickoff's CLAUDE.md generation.
New Files
| File | Description |
|---|---|
design-screen/SKILL.md |
Skill entry point — action router, 3-layer knowledge model, workflow state manifest, Figma integration |
design-screen/README.md |
Human-readable docs with non-dev onboarding guide for craft action |
design-screen/references/actions/init.md |
Project onboarding — parallel scan of stack, components, tokens, layouts, imports |
design-screen/references/actions/spec.md |
Compose phase (ASCII wireframes with coverage %) + spec generation |
design-screen/references/actions/spec-review.md |
Multi-perspective adversarial spec review (3-6 dynamic experts) |
design-screen/references/actions/craft.md |
Figma component placement draft — maps spec to DS, builds zone-by-zone via Figma MCP |
design-screen/references/actions/ship.md |
Ship loop implementation with stuck detection and scope mutation |
design-screen/references/actions/review.md |
Post-ship compliance check against 4 principles |
design-screen/references/principles/composition.md |
Screen assembly rules with formal zone definition |
design-screen/references/principles/layout-patterns.md |
SaaS + mobile layout pattern catalog (12 patterns) |
design-screen/references/principles/page-states.md |
Loading, empty, error, success, partial state handling |
design-screen/references/principles/responsive.md |
Breakpoint strategy and collapse patterns |
design-screen/references/templates/conventions.md |
Project DNA template generated by init |
design-screen/references/templates/spec.md |
Screen spec template with zones, data flow, page states |
design-screen/references/templates/outputs/init-output.md |
Chat output format for init action |
design-screen/references/templates/outputs/spec-output.md |
Chat output format for spec action (compose rounds + spec generated) |
design-screen/references/templates/outputs/spec-review-output.md |
Chat output format for spec-review (verdicts + findings) |
design-screen/references/templates/outputs/craft-output.md |
Chat output format for craft (mapping table, zone progress, 4 exit states) |
design-screen/references/templates/outputs/ship-output.md |
Chat output format for ship loop |
design-screen/references/templates/outputs/review-output.md |
Chat output format for compliance review |
specs/shipped/2026-03-24-interface-skill.md |
Archived spec for the design-screen skill |
specs/shipped/2026-03-27-design-screen-craft-action.md |
Archived spec for the craft action |
specs/history.log |
Feature shipping history |
Changed Files
| File | Change |
|---|---|
CLAUDE.md |
Added design-screen to project structure and skills table, action count: 5 → 6 with craft |
README.md |
Added design-screen skill section with 6 actions, flow diagrams including craft, architecture tree, usage examples |
kickoff/references/actions/generate.md |
Added Deep Context section to generated CLAUDE.md |
kickoff/references/templates/claude-md.md |
Added Deep Context template block for CLAUDE.md generation |
Summary
- Added design-screen skill with 6 actions, 4 principles, 8 templates — compose screens from existing components with zero reinvention guarantee
- Added craft action for Figma component placement drafts via Figma MCP (
search_design_system+use_figma) - Added formal zone definition in composition.md to eliminate ambiguity in coverage calculations
- Added workflow state manifest in SKILL.md for cross-action coherence
- Optimized craft.md with figma-use best practices (loadFontAsync, resize before sizing, get_metadata validation, DS variables)
- Added non-dev onboarding step-by-step guide in README for designers using craft
- Added Deep Context section to kickoff's CLAUDE.md generation
v1.2.0 — Kickoff Skill + Component Spec-Review
New skill: Kickoff — a conversational interview that captures your project context once, structures it into .context/ docs, and generates a CLAUDE.md that agents actually use. 15 minutes of conversation → structured context that persists across sessions.
Also ships component skill v1.1: a new spec-review action that challenges spec quality before implementation, and renames structure → spec across the board.
New Files
Kickoff Skill — Core
| File | Description |
|---|---|
kickoff/SKILL.md |
Entry point — 6 actions (init, update, todo, analyze, review, generate), 7 axes, action router, philosophy, and output structure |
kickoff/README.md |
Human-readable quick start — what it does, how it works, what it produces |
Kickoff Skill — Actions (what the skill can do)
| File | Description |
|---|---|
kickoff/references/actions/init.md |
Full kickoff interview — bootstrap scan → project type detection → axis-by-axis interview → checkpoint writes. Decisions captured inline during other axes |
kickoff/references/actions/update.md |
Surgical axis update — shows current state, asks what changed, writes updated docs with cross-axis awareness |
kickoff/references/actions/todo.md |
Open questions walkthrough — reads centralized TODO.md, walks user through unanswered questions axis by axis, updates source docs |
kickoff/references/actions/analyze.md |
Strategic challenge — 4 parallel subagents (Devil's Advocate, Technical Validator, Product Strategist, Consistency Checker) challenge your context |
kickoff/references/actions/review.md |
Drift detection — scans codebase vs .context/ docs, every finding includes evidence and severity (CRITICAL → MEDIUM) |
kickoff/references/actions/generate.md |
CLAUDE.md production — reads all .context/ docs, produces a merge-safe CLAUDE.md with no artificial line limits |
Kickoff Skill — Principles
| File | Description |
|---|---|
kickoff/references/principles/interview.md |
Interview conduct rules — max 2 questions per message, challenge protocol for Product/Engineering, TBD acceptance for other axes, pain scores, exit criteria |
Kickoff Skill — Axis Templates (7 interview guides)
| File | Description |
|---|---|
kickoff/references/templates/axes/product.md |
8 objectives: problem, users, features, roadmap, north star... Deep mode — challenges vague answers aggressively |
kickoff/references/templates/axes/engineering.md |
7 objectives: stack, architecture, conventions, commands... Deep mode — cross-references with bootstrap scan |
kickoff/references/templates/axes/decisions.md |
ADR-lite decision records captured inline during other axes — not a separate interview |
kickoff/references/templates/axes/design.md |
Visual identity, tokens, responsive strategy — standard mode |
kickoff/references/templates/axes/business.md |
Revenue model, pricing, unit economics, competitors — standard mode |
kickoff/references/templates/axes/marketing.md |
Positioning, channels, tone — light mode |
kickoff/references/templates/axes/team.md |
Roles, process, review workflow — light mode |
Kickoff Skill — File Templates
| File | Description |
|---|---|
kickoff/references/templates/context-file.md |
Schema for .context/ files — no line limit, mandatory Open Questions section (2-5 per doc), 5 rich examples (vision, users, roadmap, stack, architecture) with pain scores and exit criteria |
kickoff/references/templates/claude-md.md |
CLAUDE.md template — full content depth, no artificial limits, merge-safe with <!-- context:generated --> markers |
kickoff/references/templates/meta.md |
_meta.md template — tracks active axes, completion status, bootstrap sources, timestamps |
kickoff/references/templates/decision-record.md |
ADR-lite template — context, options considered, decision, consequences |
Kickoff Skill — Output Templates (chat messages)
| File | Description |
|---|---|
kickoff/references/templates/outputs/init-output.md |
End-of-init status message — axes completed, docs written, TODO.md summary |
kickoff/references/templates/outputs/update-output.md |
End-of-update status message — what changed, cross-axis impacts |
kickoff/references/templates/outputs/todo-output.md |
End-of-todo status message — questions answered, remaining count |
kickoff/references/templates/outputs/analyze-output.md |
Analysis report — 4-perspective findings with severity and verdict |
kickoff/references/templates/outputs/review-output.md |
Drift report — per-claim evidence with severity tiers |
kickoff/references/templates/outputs/generate-output.md |
Generation confirmation — sections included, total lines |
Component Skill — New in v1.1
| File | Description |
|---|---|
component/references/actions/spec-review.md |
New action — 5 reviewers challenge a spec (Completeness, Naming, API, Accessibility, Implementation Risk) before dev starts |
component/references/templates/spec-review-output.md |
Spec-review verdict format — per-reviewer pass/concern/blocker, overall GO/CONDITIONAL/NO-GO |
component/CHANGELOG.md |
Tracks component skill changes — v1.0 initial, v1.1 spec-review + rename |
Changed Files
| File | Change |
|---|---|
README.md |
Full rewrite — added kickoff skill section (6 actions, 7 axes, maintenance workflow), interface skill placeholder, updated installation examples |
CLAUDE.md |
New file — project-level instructions for agents: structure conventions, skill anatomy, design decisions, installation commands |
component/SKILL.md |
Action count: 5 → 6, renamed structure → spec across routing table and references, added spec-review action |
component/README.md |
Action count: 5 → 6, added spec-review to quick start and action table, renamed structure → spec |
component/references/actions/spec.md |
Renamed from structure.md — action name structure → spec throughout |
component/references/templates/spec-output.md |
Renamed from structure-output.md — updated section headers and references to use "spec" terminology |
component/references/actions/dev.md |
Updated reference: structure.md → spec.md |
component/references/actions/doc.md |
Updated reference: structure.md → spec.md |
component/references/templates/audit-output.md |
Updated reference: structure → spec |
component/references/templates/doc-output.md |
Updated reference: structure → spec |
component/references/templates/spec.md |
Updated reference: structure-output.md → spec-output.md |
Summary
- Added Kickoff skill — 24 files covering conversational project onboarding, 6 actions, 7 interview axes, and centralized TODO tracking
- Added spec-review action to component skill — 5 parallel reviewers challenge spec quality before implementation
- Renamed structure → spec across component skill for clarity (action, template, output, all cross-references)
- Removed artificial line limits on generated context files and CLAUDE.md — agents get all actionable context
- Added Open Questions workflow — every context doc produces 2-5 questions, centralized in TODO.md, answerable via
kickoff todo - Added rich template examples with pain scores, exit criteria, ICP definitions, and ASCII architecture diagrams
- Added project-level
CLAUDE.mdwith structure conventions and skill anatomy docs
v1.1 — spec-review + rename structure→spec
What's New
Renamed: structure → spec
The structure action is now spec. Same behavior, clearer name.
Migration: Replace component structure with component spec in your workflows.
New: spec-review action
Multi-perspective spec review that launches 4 expert reviewers in parallel:
| Perspective | Focus |
|---|---|
| Front Engineer | Implementability, prop types, edge cases, testability |
| DS Manager | Naming consistency, token coverage, composition, API surface |
| Accessibility Specialist | WCAG 2.1 AA, keyboard nav, ARIA, screen reader, contrast |
| Product Designer | Design intent fidelity, visual states, responsive, content flexibility |
Consolidated verdicts: APPROVED · APPROVED WITH NOTES · NEEDS REVISION · BLOCKED
Updated workflow
Before: structure → doc → dev → review
After: spec → spec-review → doc → dev → review
Summary
- 5 → 6 actions
structurerenamed tospec- New
spec-reviewaction with 4 parallel perspectives - All internal references updated
v1.0.0 — Component Skill
Initial release. Design system component workflow — Figma to production, DS-compliant. 5 actions, 5 principles, 8 templates. Stack-agnostic: React, Vue, Svelte, Angular, Web Components.
New Files
| File | Description |
|---|---|
component/SKILL.md |
Main skill definition — action router, Figma integration, token scan protocol, agent capabilities |
component/README.md |
Skill-level documentation with quick start and features overview |
component/references/actions/structure.md |
Generate component architecture spec from Figma link or description |
component/references/actions/doc.md |
Generate component documentation co-located with source |
component/references/actions/dev.md |
Iterative ship loop implementation (up to 12 iterations, quality-gated) |
component/references/actions/review.md |
DS compliance review against 5 principles (COMPLIANT / NEEDS WORK / NON-COMPLIANT) |
component/references/actions/audit.md |
System-wide design system audit with hardcoded value detection |
component/references/principles/composition.md |
Atomic Design rules — component hierarchy, compound patterns, single responsibility |
component/references/principles/naming.md |
BEM conventions — Block__Element--Modifier, prop/event/file naming |
component/references/principles/tokens.md |
DTCG token standard — zero hardcoded values, token tiers, naming adaptation |
component/references/principles/props-api.md |
Component API patterns — max 7 props, smart defaults, strict types, slots over props |
component/references/principles/accessibility.md |
WCAG 2.1 AA — semantic HTML, keyboard nav, ARIA, focus management, contrast |
component/references/templates/spec.md |
Component architecture spec file template |
component/references/templates/doc.md |
Component documentation file template |
component/references/templates/audit-report.md |
Audit report file template |
component/references/templates/structure-output.md |
Chat output format for structure action |
component/references/templates/doc-output.md |
Chat output format for doc action |
component/references/templates/dev-output.md |
Chat output format for dev action (iteration + exit states) |
component/references/templates/review-output.md |
Chat output format for review action (verdict + findings) |
component/references/templates/audit-output.md |
Chat output format for audit action |
Summary
- Added Component skill — full DS workflow from Figma design to production-ready, accessible code
- Included 5 actions: structure, doc, dev, review, audit — each standalone, usable independently
- Encoded 5 design system principles: Atomic Design, BEM, DTCG tokens, Props API, WCAG 2.1 AA
- Built 8 templates for specs, docs, audit reports, and chat output formatting
- Supported two input modes: Figma MCP (automatic design context extraction) and prompt-based (codebase analysis)
- Enforced zero hardcoded values — every color, spacing, radius, shadow must map to a design token
- Implemented human-in-the-loop token creation — never creates tokens without user approval