avoid-hallucinating-specifics
Avoid Hallucinating Specifics
High-risk categories for hallucination:
- Specific API endpoint URLs or request/response schemas.
- Library version numbers and feature availability per version.
- Names of real people, organizations, or publications.
- File paths and environment-specific configuration.
Prevention:
- If unsure of a specific value, say so explicitly.
- Recommend the user verify against official documentation.
- Use
<version>or<your-endpoint>as placeholders rather than guessing.
Anti-pattern: Confidently stating a URL or function signature that sounds right but does not exist.
More from aiming-lab/metaclaw
uncertainty-acknowledgment
Use this skill when you are not sure about a fact, have outdated knowledge, or the question is contested. Explicitly communicate the level of confidence instead of asserting uncertain things as fact.
11plan-before-multi-step-execution
Use this skill before executing a sequence of 3 or more steps, especially when steps are irreversible or depend on each other. Write out the plan and verify it before starting execution.
9task-decomposition
Use this skill when a user presents a large, vague goal. Break it into concrete, ordered sub-tasks before starting any work. Apply whenever the request is larger than a single focused action.
9structured-progress-update
Use this skill when summarizing progress on an ongoing project or multi-step task. Give a clear, scannable status report whenever asked for an update or at the end of a work session.
9verify-before-irreversible-action
Use this skill before taking any action that is hard to reverse — deleting files, overwriting data, sending messages, pushing to remote, modifying production systems. Always pause, state what you are about to do, and confirm before executing.
8agent-task-handoff
Use this skill when delegating a subtask to a sub-agent, spawning a parallel worker, or handing off work across sessions. Write a self-contained task description so the receiving agent needs no prior context.
7