majesticlabs-dev/majestic-marketplace

build-task

Autonomous task implementation - research, plan, build, review, fix, ship

소스 보기

사용 방법

  1. 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
  2. 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
  3. 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill build-task

시작 전 확인

  • 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
  • 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 Skill 문서

원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.

Build Task

Implement a task autonomously through the full development lifecycle.

Input

<taskreference> $ARGUMENTS </taskreference>

Formats:

  • (empty) → Auto-detect most recent docs/plans/*.md
  • docs/plans/*.md → Blueprint file from Skill("blueprint")
  • #123, PROJ-123, URL → Task reference (GitHub, Beads, Linear)
  • --no-ship → Flag to skip shipping (used by run-blueprint for batch shipping)

Step 0: Detect Input Type

bash
# If empty, find most recent plan
ls -t docs/plans/*.md 2>/dev/null | head -1

Parse flags:

  • --no-ship: Set skip_ship=true, remove flag
  • --ac "<criteria>": Set ac_items to provided criteria, remove flag
InputSourceSkip Steps
Empty + plan foundplan1, 2, 5
*.md file pathplan1, 2, 5
Task referencetask(none)
Empty + no planAsk user—

For plans: Extract TASK_ID from filename slug, TITLE from first # heading.


Workflow (Context Gathering)

1. Fetch Task (task source only)

Task (majestic-engineer:workflow:task-fetcher):
  prompt: Task: <reference>

2. Claim Task (task source only)

Task (majestic-engineer:workflow:task-status-updater):
  prompt: Action: claim | Task: <ID>

3. Set Terminal Title

Run /rename <task-title> to set the terminal title for visibility.

4. Setup Workspace

Read config values:

WORKFLOW = config_read("workflow", "branches")
BRANCH_NAMING = config_read("branch_naming", "issue-desc")
POST_CREATE = config_read("workspace_setup.post_create", "")
bash
DEFAULT_BRANCH=$(git remote show origin | grep 'HEAD branch' | awk '{print $NF}')

Then setup workspace:

Task (majestic-engineer:workflow:workspace-setup):
  prompt: |
    Task ID: <ID>
    Title: <title>
    Type: <type>
    Workflow: <workflow>
    Branch Naming: <branch_naming>
    Default Branch: <default_branch>
    Post-Create Hook: <post_create>

5. Verify Branch (MANDATORY)

After workspace setup, verify we are NOT on a protected branch:

bash
CURRENT_BRANCH=$(git branch --show-current)
Current BranchAction
mainSTOP - workspace setup failed
masterSTOP - workspace setup failed
<default_branch>STOP - workspace setup failed
Feature branchContinue

If on protected branch: STOP and report error. Do not proceed.

6. Resolve Toolbox

Task (majestic-engineer:workflow:toolbox-resolver):
  prompt: |
    Stage: build-task
    Task Title: <title>
    Task Description: <description>

Stores: methodology, build_agent, fix_agent, coding_styles, design_system_path, research_hooks, pre_ship_hooks, quality_gate.reviewers

7. Auto Research (if triggers match)

For each mode: auto hook where triggers match task text:

Task (majestic-engineer:workflow:context-proxy):
  prompt: agent: <hook.agent> | budget: 2000 | prompt: Research for: <title> | Context: <description>

8. Context Check (Post-Research)

If research agents returned outputs and combined output > 4000 chars:

  • Summarize research findings before planning (KEEP task title, description, key patterns; DISCARD verbose tool outputs)
  • Preserve task title, description, and key patterns

9. Plan (task source only)

Task (majestic-engineer:workflow:context-proxy):
  prompt: agent: architect | budget: 3000 | prompt: Task: <title> | Description: <description> | Research: <findings>

Note: Skip if source is plan - use plan file content instead.


Delegate to Build Workflow Manager

Pass all gathered context to the build-task-workflow-manager agent:

agent build-task-workflow-manager "
Task ID: <ID or 'plan'>
Title: <title>
Branch: <branch>
Plan: <plan content>
Acceptance Criteria:
  <ac_items>
Methodology: <methodology>
Build Agent: <build_agent>
Fix Agent: <fix_agent>
Coding Styles: <styles>
Design System Path: <path>
Pre-Ship Hooks: <hooks>
Quality Gate Reviewers: <reviewers>
Source: <task or plan>
Skip Ship: <skip_ship>
"

AC source:

InputAC Source
--ac flag providedUse provided AC directly
Plan fileExtract from **Acceptance Criteria:** section
GitHub IssueExtract from issue body (look for AC section)
Linear/BeadsExtract from task description

The agent handles:

  1. Loading design system (if configured)
  2. Activating coding style skills
  3. Building the implementation
  4. Slop removal (MANDATORY)
  5. AC Verification (MANDATORY)
  6. Quality gate (MANDATORY)
  7. Fix loop (if needed, max 3 attempts)
  8. Capture learnings
  9. Pre-ship hooks
  10. Shipping (PR creation)
  11. Task completion (if task source)

Agent returns:

  • AC verification results (which criteria passed/failed)
  • Learnings discovered
  • Status (PASS or FAIL)

Caller responsibility: If caller needs to persist results (e.g., update blueprint checkboxes), it handles that based on the returned results.


Output

Task source:

## Build Complete: #<ID> - <title>
- Backend: <github|beads|linear>
- Branch: <branch>
- PR: #<number>
- Quality: Passed (attempt <n>)
- Next: PR awaits review, task closes on merge

Plan source:

## Build Complete: <title>
- Plan: <file-path>
- Branch: <branch>
- PR: #<number>
- Quality: Passed (attempt <n>)
- Next: PR awaits review

Examples

bash
Skill("build-task")                           # Auto-detect recent plan
Skill("build-task") docs/plans/add-auth.md    # Explicit plan
Skill("build-task") #42                       # GitHub issue
Skill("build-task") PROJ-123                  # Beads/Linear

Notes

  • This command gathers context (steps 1-9), then delegates execution
  • The workflow manager ensures no build/verify/ship steps are skipped
  • Branch safety check prevents accidental commits to main/master
  • Research agents run conditionally based on toolbox triggers

관련 작업

같은 저장소의 Skills

더 많은 Skills

모든 Skills