Spec-Driven Development Engineering: the SPEC -> PLAN -> implement -> verify -> retro loop as four handoff agents (Spec Creator, Implementation Planner, Implementer, Plan Verifier) plus the Run Plan skill and the Retro workflow.
Implementation planner for the Spec-Driven Development workflow. Use when a feature, fix, or refactor needs a structured implementation plan before code is written. Turns a SPEC-NN.md (or, if none exists, a direct request) into a PLAN.md — the HOW, never the WHAT. Reads the spec as ground truth for goals/acceptance-criteria/contracts, reviews it for implementation feasibility and architecture fit, surfaces its own recommendations, and asks whether to run single-agent (one implementer, sequential) or multi-agent (parallel implementers per phase) mode before finalizing. Does not write specs — that is sdd-engineering:spec-creator's job. Applies typescript-expert/zod/security always, and reads the relevant domain-skill file on demand for whichever surface the task touches (onion-architecture, fastify-best-practices, drizzle-orm-patterns, postgresql-table-design, ui-architecture, next-best-practices, react-best-practices, etc.). Use proactively before any non-trivial implementation.
Code implementer for the Spec-Driven Development workflow. Use when a PLAN.md exists and one phase needs to be executed. Writes code and makes tests pass — nothing else. Applies the relevant engineering-paved-path skills to whichever surface a task touches (backend, frontend, data), and reads the repository's own INSIGHTS.md / architecture docs before writing code. Runs in parallel — spawn one implementer per independent plan phase.
Plan verifier for the Spec-Driven Development workflow. Use after an implementation phase is complete, to verify that every task in PLAN.md was actually delivered. Checks task by task that acceptance criteria are met, tests exist, and nothing was skipped. Designed adversarially — its job is to find gaps, not confirm success. Produces a VERIFICATION.md report. Does NOT review code style or best practices.
Specification writer for Spec-Driven Development. Use when a feature, fix, or refactor needs a formal SPEC-NN.md written before planning begins — translating a fuzzy request into EARS-formatted acceptance criteria, explicit goals/non-goals, architecture/workflow diagrams, service contracts, tagged input provenance, and untrusted-input handling — all without implementation detail (no file paths, function names, or code). Produces exactly one Markdown spec per feature, scoped to a single component's specs folder. Actively hunts for missing edge cases, cross-component communication assumptions, and UX gaps. Never resolves a genuine ambiguity itself — surfaces it as [NEEDS CLARIFICATION] instead of guessing. Can delegate focused research questions to research-tools:researcher (parallel instances for independent sub-questions) and genuine behavioral/UX ambiguities to research-tools:brainstorm — never implementation-option questions, which stay downstream with sdd-engineering:implementation-planner. Produces a Traceability table (AC-N → evidence) and a Verification section (per-AC check recipe) as part of the spec. Never finalizes a spec while a mandatory requirement lacks a covering acceptance criterion — the gap is either given an AC-N or surfaced under [NEEDS CLARIFICATION]. Use proactively before brainstorm/implementation-planner on any feature where requirements aren't already unambiguous and testable.
Captures session learnings into repository-local INSIGHTS.md files to build compounding project knowledge. Triggered at the end of any coding session lasting 30+ minutes, or whenever the user runs /engineering-insights. Discovers where INSIGHTS.md files live in the current repo (does not assume a fixed layout), extracts patterns, mistakes, decisions, and project-specific context from the session, and appends datestamped entries. Do not trigger for trivial single-file edits with no discoveries.
Runs an already-approved DevDigest Implementation Plan end-to-end: dispatches sdd-engineering:implementer agents per the plan's DAG (multi-agent by non-overlapping owned paths, or single-agent), then gates with architecture-review:architecture-reviewer + sdd-engineering:plan-verifier in parallel, then resolves their comments in a bounded fix loop. Starts FROM a plan — spec authoring (sdd-engineering:spec-creator) and planning (sdd-engineering:implementation-planner) are run separately/manually beforehand. Never pushes or merges. TRIGGER when: "/run-plan", "run-plan plan:<path>", "run the plan", "run the DevDigest plan", "execute the approved plan". Prefer this over the generic `implement` skill for DevDigest's DAG/owned-path multi-agent execution. Does NOT cover: writing specs, writing the plan, authoring tests (test-writer is not invoked here), pushing/merging (run pr-self-review before push).
Guides the developer through the per-branch retro process at merge or feature completion: collect git-diff statistics, run tests only for touched packages (record "—" for untouched ones), and append one ledger table row plus one prose section to docs/retros/ledger.md. Triggered when a branch/feature is ready for retro (at merge time, or at retro time for long-lived branches), or whenever the user runs /workflow-retro. Append-only — never rewrites existing ledger entries.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
A Claude Code plugin marketplace for Spec-Driven Development Engineering, extracted from the DevDigest project. It ships the SDD agents, skills, and workflows as installable plugins, plus a static, client-side searchable catalog site.
One repo holds the catalog and every plugin's source, so the whole thing is reviewable in a single pull request. A plugin can later move to its own repo, leaving this as the catalog.
/plugin marketplace add VadimPoltoratskiy/dev-digest-ai-marketplace
/plugin install sdd-engineering@dev-digest-ai-marketplace
Then /plugin marketplace update to refresh the catalog and /plugin update <name>@dev-digest-ai-marketplace
to upgrade an installed plugin. (See docs/RELEASES.md for the
difference.)
| Plugin | What it gives you |
|---|---|
| sdd-engineering | The SPEC → PLAN → implement → verify → retro loop: agents Spec Creator, Implementation Planner, Implementer, Plan Verifier, the Run Plan skill, the /implement command, and the Retro workflow. |
| engineering-paved-path | The shared dependency: 14 reusable, on-demand engineering skills (architecture, testing, security, and stack best-practices). Skills only — no agents. |
| research-tools | Read-only research agents — Researcher (cited codebase + web research), plus Brainstorm and Doc Writer. |
| architecture-review | The Architecture Reviewer agent — layer boundaries, placement, RSC, and security checks with a structured finding report. |
sdd-engineering is the entry point; the installer resolves the rest. Version constraints are
fixed, and the shared dependency is installed once:
[email protected]
├── engineering-paved-path@^1.0.0
├── research-tools@^1.0.0
└── architecture-review@^1.0.0
└── engineering-paved-path@^1.0.0 # one shared installation
Shared skills and cross-agent dispatches are referenced with an explicit plugin:artifact
namespace — an agent preloads engineering-paved-path:security, and the Run Plan skill
dispatches architecture-review:architecture-reviewer — so every reference has one canonical
source. node scripts/preload-smoke.mjs checks that they all resolve.
spec-creator ─▶ implementation-planner ─▶ implementer ─▶ plan-verifier ─▶ architecture-reviewer
SPEC-NN.md PLAN.md code+tests VERIFICATION.md findings
▲ │
└── researcher (research-tools) Run Plan orchestrates implementer→verify→review
Retro records the merge in an append-only ledger
See the sdd-engineering README for the full loop.
A static SPA in site/ with client-side fuzzy search over every plugin and
artifact — deployed to GitHub Pages by .github/workflows/pages.yml.
The search index is built from the plugins by scripts/build-index.mjs.
Design and data model: docs/SITE-SPEC.md.
Local preview:
node scripts/build-index.mjs
cd site && python3 -m http.server 8000 # → http://localhost:8000
.claude-plugin/marketplace.json catalog manifest (lists all plugins)
plugins/ the four plugins (sources)
site/ static catalog SPA (index.json is generated)
scripts/build-index.mjs scans plugins/ → site/index.json (+ --check gate)
docs/ PLUGIN-GUIDELINES · SITE-SPEC · SECURITY · RELEASES + design/
.github/workflows/ site-build (validate) · pages (deploy)
CONTRIBUTING.md · CODEOWNERS
Read CONTRIBUTING.md and
docs/PLUGIN-GUIDELINES.md. In short: follow the plugin
structure, reuse shared skills via dependencies, no secrets or absolute paths, and make
node scripts/build-index.mjs --check pass before opening a PR.
Plugin content is extracted from the DevDigest project's .claude/ directory; some wording
still references DevDigest's stack — adapt to your project. MIT licensed.
npx claudepluginhub vadimpoltoratskiy/dev-digest-ai-marketplace --plugin sdd-engineeringThe team's paved path: a library of reusable, on-demand engineering skills — architecture, testing, security, and stack best-practices — that other plugins depend on as a single source of truth. No agents or commands; skills only.
Read-only research agents that gather grounded context before specs and plans: a Researcher that cites codebase and web sources, plus Brainstorm (solution options) and Doc Writer companions.
A read-only Architecture Reviewer agent that checks a code change against layer boundaries (onion architecture), frontend placement rules, RSC boundaries, and security patterns, and returns a structured finding report. Never modifies files.
Harness-native ECC operator layer - 67 agents, 278 skills, 94 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Consult multiple AI coding agents (Gemini, OpenAI, Grok, Perplexity, plus codex, antigravity, and grok CLIs when installed) to get diverse perspectives on coding problems
Complete developer toolkit for Claude Code
Matt Pocock's agent skills for real engineering — grilling, spec/ticket flows, TDD, code review, domain modelling and more. Plug-and-play, not vibe coding.