smaqit-task-start

name	smaqit.task-start
description	Start working on a task. Supports autonomous mode (AI completes) or assisted mode (user approval required). Use when beginning work on tasks to set proper workflow.
metadata	{"version":"0.7.0"}

Task Start

Start working on a task with specified workflow mode: autonomous or assisted.

Usage

task.start [id]                    # Assisted mode (default) - requires user approval
task.start [id] --autonomous       # Autonomous mode - AI completes task
task.start [id] --assisted         # Explicit assisted mode

Modes

Assisted Mode (Default)

Workflow: AI implements → STOPS → User approves → User completes

• Agent implements the task
• Agent STOPS and hands back to user
• Agent MUST NOT invoke task-complete
• User reviews work and invokes /task.complete [id] if satisfied

Use for:

• Complex features requiring validation
• User-facing changes
• Changes requiring human judgment
• Quality gates before completion

Autonomous Mode

Workflow: AI implements → AI verifies → AI completes

• Agent implements the task
• Agent verifies acceptance criteria
• Agent invokes task-complete autonomously
• No user approval gate required

Use for:

• Automated workflows (CI/CD pipelines)
• Batch operations
• Well-defined tasks with clear success criteria
• Non-critical refactoring

Steps

1. Read task file (.smaqit/tasks/NNN_*.md) to understand requirements
   • If ## Findings already contains non-placeholder content, surface it in context for continuity:
     • Print: Existing findings loaded from previous execution: [summary]
2. Research map verification — check whether .smaqit/references/project-research.md exists:
   • If absent → invoke smaqit.project-research [task-id] before proceeding. Surface the resulting map in-context. Do not continue to Step 2a until the map is written.
   • If present → proceed without refreshing. The existing map is sufficient. Surface it in-context (render the table) so the implementing agent has documentation topology available. 2a. Issue triage — invoke smaqit.utils.triage-issues with the current task ID:
   • Skill reads the research map, extracts third-party tools, and searches GitHub for known open issues.
   • After triage returns, write/overwrite ## Known Issues Triage in the task file using the format from skills/smaqit.utils.triage-issues/references/TRIAGE_BLOCK.md.
   • If blocking issues found → STOP. Do not continue to Step 3. Present findings and await user direction (proceed, reframe scope, or mark as Blocked).
   • If advisory or clear → continue to Step 3. Advisory findings are visible in-context but do not require approval.
   • If triage exits cleanly (skip flag, no tools, gh unavailable, registry missing) → continue to Step 3 silently.
   • If triage write-back fails, report a warning and continue (non-blocking).
3. Determine mode from command arguments (default: assisted)
4. Update task status to "In Progress"
5. Store mode in task file as metadata field:

   **Mode:** Autonomous | Assisted
   

6. Update PLANNING.md to reflect "In Progress" status
7. If a persistent, cross-session memory/notes capability is available in this environment, use it to record task state (best-effort — PLANNING.md and the task file remain the source of truth regardless):
   • subject: "task state"
   • fact: "[NNN] [Title] — In Progress ([Assisted|Autonomous], started YYYY-MM-DD)" (≤ 200 chars)
   • citations: path to the task file (e.g., .smaqit/tasks/NNN_task_title.md)
   • reason: "Ensures in-progress task and mode are visible in any branch, supporting parallel agent workflows"
8. Load workflow rules by reading references/RULES.md
9. Begin implementation following task requirements

Task File Format

See .smaqit/templates/task.template.md for the canonical task file structure.

This skill adds the Mode field (set to Autonomous or Assisted) and the Started field (set to today's date) when starting a task.

Critical Rules

⚠️ Read references/RULES.md for complete workflow enforcement rules

For Assisted Mode:

• Agent MUST NOT complete the task autonomously
• Agent MUST stop after implementation and hand back to user
• Only user can invoke /task.complete [id]

For Autonomous Mode:

• Agent MUST verify all acceptance criteria before completing
• Agent invokes task-complete after verification
• Agent should document completion rationale

Examples

Starting an Assisted Task

User: /task.start 003
Agent: [reads task 003, sets mode to Assisted, updates status]
Agent: [implements the task]
Agent: "Task 003 implementation complete. Please review and run /task.complete 003 when satisfied."

Starting an Autonomous Task

User: /task.start 005 --autonomous
Agent: [reads task 005, sets mode to Autonomous, updates status]
Agent: [implements the task]
Agent: [verifies criteria]
Agent: [invokes task-complete 005]
Agent: "Task 005 completed autonomously. All criteria verified."