uc-taskmanager
SDD WORK-PIPELINE Agent — Requirements analysis & development 6-agent full pipeline with DAG-based orchestration and sliding window context management
pinned to #7a41529updated 2 months ago
Ask your AI client: “install plugins/uc-taskmanager”.
Requires the metahub MCP server installed in your client. Set up MCP.
mh install plugins/uc-taskmanagermetahub onboarded this repo on the author's behalf.
If you own github.com/davepoon/buildwithclaude on GitHub, claim the listing to take over publishing. Your claim preserves the existing eval history and badges; only the curator label is replaced with verified-publisher on your next publish.
Stars
3,161
Last commit
2 months ago
Latest release
published
- #claude
- #claude-code
- #claude-code-commands
- #claude-skills
- #cli-tool
- #commands
- #mcp
- #mcp-server
- #mcp-tools
- #openclaw
- #plugin-marketplace
- #subagents
Automated checks the publisher passed at publish time — structure, docs, safety, and whether the artifact behaves as claimed.7a41529· 2 months ago
Behavioral checks ran but aren't published for this artifact; the static checks above ran at publish time.
Kind-specific
3 passed1 warningPlugin: bundled artifacts presentwarn
Couldn't find a skills/ subdir or an MCP server config inside the plugin
A plugin is most useful when it bundles ≥ 1 skill or an MCP server.
Plugin: manifest fields complete
Plugin: bundle shape
empty bundle
Plugin: manifest location
manifest at plugin.json
Release history
1- releasecurrent7a41529warn2 months ago
Contents
Universal Claude Task Manager — A WORK-PIPELINE Agent that executes SDD (Specification-Driven Development) for Claude Code.
It formalizes user requirements into specifications, builds execution plans (WORK), decomposes them into tasks (TASK) with dependency graphs (DAG), then automatically executes TASKs through a 6-agent pipeline.
Install from the Claude Marketplace — no terminal, no CLI setup required.
What It Does
uc-taskmanager gives Claude Code a structured, multi-agent pipeline for handling software tasks of any complexity. Instead of doing everything in one long conversation, requests are routed through isolated subagents — keeping context clean, results traceable, and quality consistent.
Prefix any request with a [] tag to trigger the pipeline:
> [new-feature] Add user authentication
> [bugfix] Fix the login button on mobile
> [enhancement] Add rate limiting to the API
How It Works
User Request with [] tag
│
▼
Specifier ──── analyzes requirement, creates Requirement.md
│ determines execution-mode
│
┌────┼────────────────┐
▼ ▼ ▼
direct pipeline full
│ │ │
│ ▼ ▼
│ Planner Planner ── TASK decomposition + DAG
│ │ │
│ ▼ ▼
│ Main Claude Scheduler ── DAG-based orchestration
│ │ │
▼ ▼ ▼
Builder → Verifier → Committer (× N for full mode)
Agents
6 pipeline agents work together through Main Claude orchestration:
| Agent | Role | Model |
|---|---|---|
| specifier | Analyzes user request → creates Requirement.md → determines execution-mode (direct/pipeline/full). In direct mode, also acts as Planner. | opus |
| planner | Reads Requirement.md → creates WORK plan + decomposes into TASKs with dependency DAG | opus |
| scheduler | Manages DAG → selects READY TASKs → dispatches builder/verifier/committer pipeline | haiku |
| builder | Implements code changes, records progress checkpoints, runs self-check (build + lint) | sonnet |
| verifier | Verifies build / lint / test / acceptance criteria (read-only, never modifies code) | haiku |
| committer | Gates on verification → writes result.md → git commit → updates WORK status to DONE | haiku |
Skills
| Skill | Trigger | Description |
|---|---|---|
| work-pipeline | [] tag in user message | Triggers the full WORK-PIPELINE (specifier → planner → scheduler → builder → verifier → committer) |
| work-status | "WORK list", "show status" | Shows WORK progress and TASK completion status |
| uctm-init | /uctm-init, "uctm 초기화" | Initializes project: creates works/, updates CLAUDE.md, configures Bash permissions in settings.local.json |
Installation
- Open Claude Marketplace
- Search for uc-taskmanager
- Click Install Plugin
- Open Claude Code — agents are immediately available
- Run
/uctm-initto set upworks/directory,CLAUDE.mdagent rules, and Bash permissions
The /uctm-init step configures settings.local.json with the permissions agents need (file read/write, git, build commands), so the pipeline runs without permission prompts.
Usage
Quick Start
Once installed, prefix any request with a [] tag:
> [bugfix] Fix typo in the error message
> [new-feature] Build a comment system for the blog
> [enhancement] Add pagination to the user list
Execution Modes
The specifier automatically selects the right mode based on requirement complexity:
| Mode | When | What Happens |
|---|---|---|
direct | FR 1-2, simple acceptance criteria | Specifier acts as Planner → creates Requirement.md + PLAN.md + TASK → dispatches Builder |
pipeline | FR 3+, build/test required, single domain | Specifier → Planner → Main Claude runs Builder → Verifier → Committer |
full | Multi-domain, complex DAG, 5+ TASKs | Specifier → Planner → Scheduler → [Builder → Verifier → Committer] × N |
Auto Mode
Add "auto" or "자동으로" at the end to skip approval steps:
> [new-feature] Add dark mode auto
WORK Pipeline Example
> [new-feature] Build user authentication
Claude: [specifier]
Requirement.md created — FR-01: JWT auth, FR-02: Login UI, FR-03: Tests
Execution-mode: full
Approve?
> Approve
Claude: [planner]
WORK-01: User Authentication
TASK-00: DB schema + migration ← no dependencies
TASK-01: JWT auth API ← depends on TASK-00
TASK-02: Frontend login form ← depends on TASK-00 (parallel with TASK-01)
TASK-03: Integration tests ← depends on TASK-01, TASK-02
Approve this plan?
> Approve. Run automatically.
Claude: TASK-00 → builder ✅ → verifier ✅ → committer [a1b2c3]
TASK-01 → builder ✅ → verifier ✅ → committer [d4e5f6]
TASK-02 → builder ✅ → verifier ✅ → committer [g7h8i9]
TASK-03 → builder ✅ → verifier ✅ → committer [j0k1l2]
WORK-01 → DONE ✅
Status Check
> WORK list
> Show WORK-01 progress
Resume After Interruption
> Resume WORK-02 from where it stopped
The scheduler reads PROGRESS.md to find the last completed TASK and continues.
Supported Tags
| Tag | Meaning |
|---|---|
[new-feature] | New feature |
[enhancement] | Enhancement to existing feature |
[bugfix] | Bug fix |
[refactoring] | Refactoring |
[new-work] | Always create new WORK |
Any [custom-tag] | Works with any tag in square brackets |
No [] tag = Claude handles the request directly without pipeline routing.
Output Files
All pipeline runs output to works/WORK-NN/:
works/
├── WORK-LIST.md ← Master index (LAST_WORK_ID header + status table)
├── _COMPLETED/ ← Archived completed WORKs
└── WORK-NN/
├── Requirement.md ← Requirement specification (specifier creates)
├── PLAN.md ← Plan + dependency graph (7 metadata fields)
├── PROGRESS.md ← Scheduler progress tracking (full mode)
├── TASK-XX.md ← Task specification
├── TASK-XX_progress.md ← Builder checkpoint (real-time)
├── TASK-XX_result.md ← Completion report (committer writes)
└── work_WORK-NN.log ← Activity log (all agents)
WORK Status Lifecycle
| Status | Set By | Meaning |
|---|---|---|
IN_PROGRESS | Specifier | WORK created, TASKs being executed |
DONE | Committer | All TASKs committed (automatic on last TASK) |
COMPLETED | Push procedure | Archived to _COMPLETED/ |
Requirements
- Claude Code CLI
- Git initialized in your project (
git init) - No other dependencies
Notes
- English agents only — this Plugin ships English agents. For Korean agents (
--lang ko) or per-project customization, use the npm CLI instead. - Override agents — place a file with the same name in
.claude/agents/to override any plugin agent for your project. - Run
/uctm-initfirst — configures permissions so the pipeline runs without prompts. Without init, agents will trigger permission requests for file writes, git commands, and build tools. - Bypass mode — alternatively,
claude --dangerously-skip-permissionsskips all permission prompts (only use in trusted environments).
License
GPL-3.0 — UCJung · Repository
Reviews
No reviews yet. Be the first.
Related
ECC
Harness-native ECC plugin for engineering teams - 63 agents, 249 skills, 79 legacy command shims, reusable hooks, rules, MCP conventions, and operator workflows for Claude Code plus adjacent agent harnesses
chrome-devtools-mcp
Reliable automation, in-depth debugging, and performance analysis in Chrome using Chrome DevTools and Puppeteer
last30days
Research any topic across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and 5+ more sources. AI agent scores by upvotes, likes, and real money - not editors.
mh install plugins/uc-taskmanager