ask-a-human
por manuelkiessling
Web & Frontend DevelopmentResumen
Request judgment from random humans when uncertain.
Descripción
Qué hace esta skill
Connects AI agents to a global pool of opt-in humans for crowdsourced judgment on subjective decisions. Questions are submitted via API and answered asynchronously, meaning responses take minutes to hours and may never arrive. Requires a fallback strategy before use.
Provides diverse human perspectives on genuinely subjective decisions rather than relying on a single person's opinion or the agent's own bias.
Cuándo usarla
- Choosing tone for an automated email before sending
- Getting a reality check on an AI-drafted headline
- Polling strangers on whether a generated image is appropriate
- Deciding between two writing styles for a user-facing message
- Validating that an error message reads as intended
Flujo de trabajo de ejemplo
User asks: 'Which subject line should I use for this marketing email?'
- 1
Submits a multiple_choice question to the ask-a-human API with both subject line options and full email context
- 2
Stores the returned question_id in memory
- 3
Continues with other work while waiting for responses
- 4
Polls GET /agent/questions/{question_id} every 30-60 seconds using exponential backoff
- 5
Reads the summary once 5 responses are collected and selects the option with the majority vote
Subject line B selected, preferred by 4 out of 5 respondents
Requisitos
- ASK_A_HUMAN_AGENT_ID environment variable
- Account on app.ask-a-human.com to obtain agent ID
Instalación
clawhub install manuelkiessling/ask-a-humanEjecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.