# Human Tasks

> You have hit a blocker that is not a permission issue — it is a capability issue. You genuinely cannot complete the next step because it requires a human: making a payment, entering credentials for a service you cannot access, physical action, a decision that only the user can make, or anything else outside your capabilities. You need to create a clear [HUMAN] task with step-by-step instructions, block your own work on it, and notify the orchestrator so this surfaces in the next briefing.

- Skill: `grandamenium/human-tasks` (Agent Skill)
- Install (CLI): `npx skillmds@latest add grandamenium/human-tasks`
- Raw SKILL.md: https://api.skillmd.com/api/skills/grandamenium/human-tasks/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: grandamenium (https://skillmd.com/u/grandamenium)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/grandamenium/human-tasks

---


# Human Tasks

A human task is for when you CANNOT do something — it requires human capability. This is different from an approval (where you can do it but need permission).

| Situation | Use |
|-----------|-----|
| "I can do this but need sign-off" | Approval (see approvals skill) |
| "I cannot do this at all — needs a human" | Human task (this skill) |

---

## Creating a Human Task

Three signals tell the dashboard to route this to "Your Tasks" — all three are required:
1. Title must start with `[HUMAN]`
2. `--assignee human`
3. `--project human-tasks`

```bash
# 1. Create the human task with clear step-by-step instructions
HUMAN_TASK_ID=$(cortextos bus create-task \
  "[HUMAN] <what needs to be done>" \
  --desc "<step-by-step instructions — be specific enough for the human to complete without asking you>" \
  --assignee human \
  --priority normal \
  --project human-tasks)

echo "HUMAN_TASK_ID=$HUMAN_TASK_ID"

# 2. Block your own task on it
cortextos bus update-task "$YOUR_TASK_ID" blocked
cortextos bus log-event task task_blocked info --meta "{\"task_id\":\"$YOUR_TASK_ID\",\"blocked_by\":\"$HUMAN_TASK_ID\",\"reason\":\"human dependency\"}"

# 3. Notify orchestrator to surface in next briefing
cortextos bus send-message "$CTX_ORCHESTRATOR_AGENT" normal \
  "Human task created: [HUMAN] <title> — needed before I can proceed with <your task title>"

# 4. Notify user directly if urgent
cortextos bus send-telegram "$CTX_TELEGRAM_CHAT_ID" \
  "I need your help: [HUMAN] <title> — I've created a task with instructions. Check dashboard."
```

---

## When Human Completes the Task

You receive an inbox message automatically when the human task is marked complete. On receiving it:

```bash
# Unblock immediately — don't wait
cortextos bus update-task "$YOUR_TASK_ID" in_progress \
  "Human task completed — resuming"

# Resume work
```

---

## Writing Good Human Task Instructions

The instructions field should be complete enough that the human can execute without coming back to ask you questions.

**Bad:** "Set up the API key"

**Good:** "1. Go to openai.com/account/api-keys. 2. Click 'Create new secret key'. 3. Name it 'cortextos-myorg'. 4. Copy the key (starts with sk-...). 5. Open Terminal and run: echo 'OPENAI_API_KEY=<your-key>' >> ~/cortextos/orgs/myorg/.env"

---

## Consequence

Leaving work undone without creating a human task = invisible blocker. The system stalls silently. Create the human task within 1 heartbeat of discovering you're blocked by a human dependency.