autodev-run

name	autodev-run
description	Use when the user wants to dispatch a coding task to a remote workspace — fixing bugs, implementing features, running an agentic coding run, or creating a PR automatically. Use this even if the user just says "fix the login bug on myproject" or "run this task" or "create a run" or "implement X on Y". Also use when the user asks to approve or reject a pending run.

Create & Monitor Run

Create a coding task run on the AgenticKode platform and monitor it to completion.

API Base: http://localhost:8000/api

Usage

/autodev-run <task-description> — Interactive: list projects, let user pick, create run /autodev-run <project-slug> <task-description> — Direct: create run on specific project

Workflow

Step 1: Resolve Project

If no project specified, list available projects and ask user to pick:

curl -s http://localhost:8000/api/projects | python3 -c "
import json, sys
projects = json.load(sys.stdin)
for p in projects:
    ws = len(p.get('workspace_server_ids', []))
    print(f\"  {p['project_slug']:30s} {p['repo_owner']}/{p['repo_name']}  ({ws} workspace{'s' if ws != 1 else ''})\")"

Ask user to pick a project by slug name.

Step 2: Create the Run

curl -s -X POST http://localhost:8000/api/runs \
  -H "Content-Type: application/json" \
  -d '{
    "project_id": "<PROJECT_ID>",
    "title": "<TASK_DESCRIPTION>",
    "description": "<EXPANDED_DESCRIPTION>"
  }' | python3 -c "import json,sys; r=json.load(sys.stdin); print(f'Run #{r[\"id\"]} created — status: {r[\"status\"]}')"

Replace <PROJECT_ID> with the project's project_id, and fill in the title/description from the user's task.

Step 3: Monitor Progress

Poll every 15 seconds until the run reaches a terminal state (completed, failed, cancelled) or awaiting_approval:

curl -s http://localhost:8000/api/runs/<RUN_ID> | python3 -c "
import json, sys
r = json.load(sys.stdin)
phase = r.get('current_phase', 'unknown')
status = r['status']
print(f'Run #{r[\"id\"]}: {status} (phase: {phase})')
if r.get('error_message'):
    print(f'  Error: {r[\"error_message\"][:200]}')
"

Report each phase transition to the user. Terminal states:

• completed — Report success, show PR link if available
• failed — Report error message and which phase failed
• awaiting_approval — Tell user a PR is ready for review, offer to approve/reject

Step 4: Handle Approval (if needed)

If the run reaches awaiting_approval:

Approve:

curl -s -X POST http://localhost:8000/api/runs/<RUN_ID>/approve \
  -H "Content-Type: application/json" \
  -d '{"merge": true}'

Reject:

curl -s -X POST http://localhost:8000/api/runs/<RUN_ID>/reject \
  -H "Content-Type: application/json" \
  -d '{"reason": "<REASON>"}'

Step 5: Get Results

Once completed, fetch the final run details including logs:

curl -s http://localhost:8000/api/runs/<RUN_ID>/logs | python3 -c "
import json, sys
logs = json.load(sys.stdin)
for entry in logs[-20:]:
    print(f'[{entry.get(\"phase\",\"\")}] {entry.get(\"message\",\"\")[:150]}')
"

Notes

• Polling interval: 15 seconds for active runs
• Report each phase change to the user as it happens
• If a run takes more than 10 minutes, ask the user if they want to keep waiting
• The default pipeline steps are: workspace_setup, init, planning, coding, testing, reviewing, approval, finalization (flow prompts replace these with a single agent call — ADR-009)