How to use
- Copy the installation command and review the source package before installing it in your own agent environment.
- Check the source document for required tools, credentials and supported hosts.
- Describe your input and desired deliverable; review the output against the source material.
npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill build-taskBefore you start
- Installation does not include a model subscription, API allowance or permission to access your files.
- Repository stars and installation counts describe activity, not an independent quality or security test.
Rendered from the source repository. Headings, examples, code, tables, links, and referenced images are preserved.
Build Task
Implement a task autonomously through the full development lifecycle.
Input
<taskreference> $ARGUMENTS </taskreference>
Formats:
- (empty) → Auto-detect most recent
docs/plans/*.md docs/plans/*.md→ Blueprint file fromSkill("blueprint")#123,PROJ-123, URL → Task reference (GitHub, Beads, Linear)--no-ship→ Flag to skip shipping (used by run-blueprint for batch shipping)
Step 0: Detect Input Type
# If empty, find most recent plan
ls -t docs/plans/*.md 2>/dev/null | head -1Parse flags:
--no-ship: Setskip_ship=true, remove flag--ac "<criteria>": Setac_itemsto provided criteria, remove flag
| Input | Source | Skip Steps |
|---|---|---|
| Empty + plan found | plan | 1, 2, 5 |
*.md file path | plan | 1, 2, 5 |
| Task reference | task | (none) |
| Empty + no plan | Ask user | — |
For plans: Extract TASK_ID from filename slug, TITLE from first # heading.
Workflow (Context Gathering)
1. Fetch Task (task source only)
Task (majestic-engineer:workflow:task-fetcher):
prompt: Task: <reference>2. Claim Task (task source only)
Task (majestic-engineer:workflow:task-status-updater):
prompt: Action: claim | Task: <ID>3. Set Terminal Title
Run /rename <task-title> to set the terminal title for visibility.
4. Setup Workspace
Read config values:
WORKFLOW = config_read("workflow", "branches")
BRANCH_NAMING = config_read("branch_naming", "issue-desc")
POST_CREATE = config_read("workspace_setup.post_create", "")DEFAULT_BRANCH=$(git remote show origin | grep 'HEAD branch' | awk '{print $NF}')Then setup workspace:
Task (majestic-engineer:workflow:workspace-setup):
prompt: |
Task ID: <ID>
Title: <title>
Type: <type>
Workflow: <workflow>
Branch Naming: <branch_naming>
Default Branch: <default_branch>
Post-Create Hook: <post_create>5. Verify Branch (MANDATORY)
After workspace setup, verify we are NOT on a protected branch:
CURRENT_BRANCH=$(git branch --show-current)| Current Branch | Action |
|---|---|
main | STOP - workspace setup failed |
master | STOP - workspace setup failed |
<default_branch> | STOP - workspace setup failed |
| Feature branch | Continue |
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:
| Input | AC Source |
|---|---|
--ac flag provided | Use provided AC directly |
| Plan file | Extract from **Acceptance Criteria:** section |
| GitHub Issue | Extract from issue body (look for AC section) |
| Linear/Beads | Extract from task description |
The agent handles:
- Loading design system (if configured)
- Activating coding style skills
- Building the implementation
- Slop removal (MANDATORY)
- AC Verification (MANDATORY)
- Quality gate (MANDATORY)
- Fix loop (if needed, max 3 attempts)
- Capture learnings
- Pre-ship hooks
- Shipping (PR creation)
- 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 mergePlan source:
## Build Complete: <title>
- Plan: <file-path>
- Branch: <branch>
- PR: #<number>
- Quality: Passed (attempt <n>)
- Next: PR awaits reviewExamples
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/LinearNotes
- 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


