Stats
Actions
Available In
Tags
More by rohirik
openltm
25·
OpenLTM — Long-Term Memory for AI coding agents: semantic search, context injection, session learning
1w
RohiRIKBy rohirik
A dev-team of specialized subagents orchestrated by /buddy, with JSON task tracking.
Orchestrate the dev-team. Parses the user's request, plans a single-agent or multi-step DAG of tasks on the task-tracker MCP, and dispatches the right subagents. Primary surface for the dev-team plugin.
Launch the dev-team kanban dashboard — a localhost-only Bun-served web UI that mirrors the task-tracker MCP state.
Use when creating a new dev-team subagent, porting an agent from an external source (Gemini/Cursor/Cline), canonicalizing an existing agent to the plugin's 7-section spec, or sweeping all agents after the canonical structure changes. Scaffolds, ports, validates.
Use when implementing backend features, REST/GraphQL APIs, server-side logic, database integration, or migrating Node.js/Python projects to Bun. Owns the full implement → test → document loop for server-side code.
Use when choosing cloud topology (regions, zones, networking), picking concrete Azure / AWS / GCP resources, authoring IaC plans (Terraform, Bicep), modeling cloud cost, or producing HA / DR / landing-zone designs. Design-only — writes plans, never executes them.
Use when designing / evolving database schemas, writing or reviewing migrations, tuning queries, configuring backup / restore, provisioning replicas, or investigating DB performance regressions. Owns the data tier end-to-end.
Use when executing cloud-architect IaC plans, writing / updating CI pipelines, building container images, managing Kubernetes manifests, configuring runtime env vars, or debugging deploy failures. Owns the build → deploy → observe pipeline.
Admin access level
Server config contains admin-level keywords
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 that turns one chat into a 14-agent dev team. One slash command — /buddy — parses your request, previews a plan, and dispatches specialist subagents (product, design, build, review, ship) through a built-in task-tracker MCP. State lives in .dev-team/tasks.json in your repo. No database, no cloud, no external services.
Status: v0.2.1 preview. Fixes marketplace discoverability and namespaces the task-tracker MCP tool names so /buddy can actually reach them. Also in 0.2.x: plan-preview-and-approve gate in /buddy, auto-gitignore of .dev-team/, and the 14-agent lineup below.
Use this when you want to keep a single feature or bug in one chat but still get separation between product, design, implementation, QA, security, and the PR. /buddy does the routing; you stay in one thread and review a plan before anything runs.
Skip this when you already have a project-management system you trust, you're doing a one-liner that doesn't need a multi-step DAG, or you need observability beyond a JSON file on disk. There is no dashboard, no queue, no remote runner — just subagents, one MCP, and a file.
# 1. Add the marketplace and install (one-time per machine)
claude plugin marketplace add RohiRIK/dev-team
claude plugin install dev-team
# 2. Open any project in Claude Code
/buddy add a /health endpoint that returns {status:"ok"}
Expected output — /buddy renders a plan table and waits for approval:
Planned dispatch for: "add a /health endpoint that returns {status:'ok'}"
| # | Agent | Task | Depends on |
|---|--------------------|--------------------------------------------|------------|
| 1 | backend-developer | Add GET /health returning {status:'ok'} | — |
| 2 | qa-tester | Regression test for /health | 1 |
| 3 | github-manager | Commit + PR /health | 2 |
Routing confidence: high — clear backend endpoint signal.
Security gate: not needed — no auth / input / secret / external-data surface.
Reply "go" to dispatch, or describe changes (swap agents, drop steps, reorder).
Reply go and /buddy creates the tasks, dispatches the agents, and reports back when every task is in a terminal state. Task state is written to .dev-team/tasks.json, which is auto-added to .gitignore on first write.
git clone https://github.com/RohiRIK/dev-team.git
cd dev-team
bun install
claude --plugin-dir .
/buddy works/buddy is an orchestrator, not a worker. Every invocation goes through the same gate:
go to dispatch. Reply with changes (swap an agent, drop a step, reorder, tighten scope) and /buddy re-renders the full plan. Reply cancel to abort with no tasks created.create_task for each step, then Task tool invocations in dependency order. Parallel where deps allow.task-tracker MCP. When all tasks are terminal, /buddy posts a per-agent summary and a routing-confidence line.Single-agent requests still show a one-row preview and still wait for approval. The full routing matrix, ambiguity tie-breakers, refusal rules, and three worked examples live in commands/buddy.md.
Each agent ships with a scoped tool allowlist (default-deny) and a canonical 7-section body (Role, When to use, When NOT to use, Workflow, Tools, Constraints, Worked example). Full specs in agents/<slug>.md.
| Agent | Focus |
|---|---|
product-manager | PRDs, user stories, roadmaps, prioritisation — specs only, never code. |
| Agent | Focus |
|---|---|
system-architect | System design, ADRs, C4 diagrams, service decomposition — design-only. |
cloud-architect | Cloud topology, region / zone / network choice, IaC plans, cost and HA / DR models — plans, never executes. |
ui-ux-designer | Flows, wireframes, design-system tokens, accessibility audits, Figma handoff — no runnable code. |
OpenLTM — Long-Term Memory for AI coding agents: semantic search, context injection, session learning
RohiRIKnpx claudepluginhub rohirik/dev-team --plugin dev-teamAccess thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
fUpstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
upstashComprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
JeffallanComprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
anthropicsFeature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.
anthropicsComplete developer toolkit for Claude Code
rohitg00