{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreieb3i2yijwvvnlon4srh227jlmwswlkspw3togttyqseqephs6g4m",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mnvfo4vwjjw2"
  },
  "path": "/t/if-unsure-ask-never-guess-ai-agent-pre-execution-checklist/176632#post_3",
  "publishedAt": "2026-06-09T22:15:29.000Z",
  "site": "https://discuss.huggingface.co",
  "textContent": "From an MCP perspective, this document is a proposal:\n\nLet Providers supply a pre-execution Checklist per Action, in natural language.\n\nIf MCP makes an Action callable, the Provider Checklist defines the questions that must be answered before that Action becomes executable.\n\nIn the API era, describing the call format was enough.\nIn the Agent era, defining the pre-execution conditions becomes equally necessary.\n\nThe flow is simple:\n\nProvider supplies the question list.\nAgent checks whether the answers exist.\nIf not, Agent asks the user.\nIf still unanswered, Agent does not execute.\n\nIf a Provider does not supply the necessary verification items for accuracy and safety, that absence may become relevant in determining the Provider’s scope of responsibility in a post-incident review.\n\nAnd if an Agent executes without user confirmation, that execution should not be easily treated as the user’s approved decision.",
  "title": "If unsure, ask. Never guess. — AI Agent Pre-Execution Checklist"
}