Claude Code Plugins

Personal Claude Code plugin marketplace with reusable dev workflow agents.

Installation

# Add the marketplace (once per machine)
/plugin marketplace add ddehart/claude-code-plugins

# Install the dev-workflow plugin
/plugin install dev-workflow@ddehart-plugins

Available Plugins

dev-workflow

Dev workflow agents and skills for test running, branch creation, conventional commits, pull request management, Linear issue management, browser verification, and spec writing.

Agents included:

Agent	Description
test-runner	Execute test suites (unit, E2E, or all) and analyze results
branch-creator	Create properly named Git branches following conventions
commit-creator	Create commits following Conventional Commits standard
pr-manager	Manage pull request workflows (create, monitor checks, review feedback, merge)
linear-issues	Manage Linear issues (create, update, view, search, status transitions)
chrome-verifier	Verify deployed features via browser automation, returning concise summaries

Skills included:

Skill	Description
spec-writing	Interview users to refine rough ideas into comprehensive, implementation-ready specifications

meta-claude

Claude Code productivity skills for session management, self-documentation, and self-reflection.

Skills included:

Skill	Description
session-chronicle	Two-layered session journal: functional chronicle (decisions, learnings, failed approaches) and open-ended reflection on functional states. Auto-triggers deep practice review after 10+ entries without evolution. Manual /session-chronicle reflect now reviews all entries since last evolution (capped at 20).
session-export	Export Claude Code session transcripts from JSONL logs to readable text format
self-documentation	Explain Claude Code features, capabilities, and tools; record observations about undocumented behaviors

pm-workflow

Product management workflow skills for writing and evaluating PRDs and product one-pagers.

Skills included:

Skill	Description
prd-writing	Interview users to transform rough product ideas into comprehensive one-pagers for product kick-offs
prd-evaluation	Evaluate PRDs from a product leader's perspective, surfacing gaps in reasoning and unstated assumptions

knowledge-commons

Generator for per-project knowledge graphs with cross-domain promotion. Scaffolds a graph (atlas, maps, type directories), a ledgered /process pipeline, and project-owned skills from a config-driven interview; portable claims promote into a personal commons, and steering-grade principles graduate into always-loaded rules. Generated skills are owned by their project and expected to diverge, so template fixes reach them by semantic patch rather than regeneration. No validators or enforcement machinery — plan approval is the gate.

Skills included:

Skill	Description
graph-init	Interview → .commons.yml + graph scaffold + generated project-owned process and knowledge-graph skills; --config-only wires an existing hand-built graph into promotion
promote	Derive a portable claim into a target graph (or a rule into the instruction tier) with association against the target's maps and per-promotion approval
graph-patch	Propagate template fixes into a project's already-generated, hand-sharpened skills without regenerating them — applies each pending change by judgment, verifies it landed, and records what was applied

Usage

Once installed, the agents and skills are automatically available. Claude will proactively use them when you:

dev-workflow:

• Ask to run tests
• Ask to create a branch
• Ask to commit changes
• Ask to create, monitor, or merge pull requests
• Ask to manage Linear issues
• Ask to verify a deployment or check a live site
• Ask to write, create, or refine a spec

meta-claude:

• Ask about Claude Code features ("How do skills work?", "What can you do?")
• Ask decision questions ("Should I use a skill or slash command?")
• Ask to chronicle a session, write a journal entry, or reflect on what was accomplished
• Ask to export a session transcript
• Report discovering undocumented behavior

pm-workflow:

• Ask to write a PRD or product brief
• Ask to create a one-pager or kick-off doc
• Ask to define product requirements
• Ask to evaluate, review, or critique a PRD

knowledge-commons:

• Ask to set up a knowledge graph in a project, or wire an existing one into promotion
• Ask to process a chronicle entry, transcript, or other source into the graph
• Ask whether something generalizes, or to promote a claim into the commons
• Ask to graduate a principle into an always-loaded rule
• Ask what template fixes a project's generated skills are still missing

Updating

/plugin update dev-workflow@ddehart-plugins
/plugin update meta-claude@ddehart-plugins
/plugin update pm-workflow@ddehart-plugins
/plugin update knowledge-commons@ddehart-plugins

Naming Conventions

Skills use specific, intention-revealing names:

• Ask: "Reading this name in a / menu, would I know what it does?"
• Examples: spec-writing, self-documentation, session-export, graph-init, promote
• Gerunds read well and are common here, but they're a house preference, not a requirement — a skill's slash command comes from its directory name and auto-invocation is driven by its description, so the name's real job is menu discoverability. Plugin namespacing (knowledge-commons:promote) supplies disambiguation for otherwise-generic names.

Agents use role-oriented names describing their function:

• Examples: test-runner, branch-creator, pr-manager, chrome-verifier
• Pattern: Noun-noun compounds representing a specialized role

This distinction reflects their nature: skills teach expertise (something you're "adept at"), while agents perform tasks (roles that "run" or "manage" things).