comprehend-contract

Guides human comprehension checks before Loopbench contract acceptance and human gate decisions.

1|Updated Jul 30, 2026
One-click install
npx skills add https://github.com/hideh1231/loopbench --skill comprehend-contract-hideh1231
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: comprehend-contract
Source: https://github.com/hideh1231/loopbench/tree/main/skills/comprehend-contract
Command: npx skills add https://github.com/hideh1231/loopbench --skill comprehend-contract-hideh1231

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It prevents humans from approving Loopbench contracts or human gates without understanding the decision, by enforcing a structured explain-then-verify workflow before any acceptance is recorded. ## Core Features & Use Cases - Bounded comprehension packets: Reads loopbench comprehension packet output as the source of truth for the decision's purpose, choices, evidence, and limits. - Learning modes: Follows the contract's declared mode — off (no questions), light (one explicit confirmation), or deep (decision-mapped quiz with grading and score recomputation). - Record integrity: Stores memos and quizzes via loopbench comprehension record, refuses to infer ambiguous answers, and supports comprehension invalidate to revoke records made without a real human answer. - Use Case: Before accepting a contract that changes a paid external service, the skill explains the tradeoffs, asks decision-relevant questions, grades the answers, and records a verified pass. ## Quick Start Ask the AI to run the comprehend-contract workflow for a task by reading its comprehension packet and walking you through the explanation and confirmation before acceptance.

Frequently Asked Questions about comprehend-contract

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify human understanding before accepting a Loopbench contract?

Run `loopbench comprehension packet <task-id>` to read the bounded packet, explain the purpose, choices, evidence, and limits, then follow the contract's declared learning mode. Record the result with `loopbench comprehension record` including the memo and quiz files.

What are the learning modes off, light, and deep in Loopbench comprehension?

Off skips all understanding questions and records nothing. Light asks one explicit confirmation question and copies the verbatim answer into the record note. Deep requires decision-mapped questions, a grade table, and a recomputed score before recording a pass.

Can a comprehension record replace external verification or acceptance?

No. An understanding record is not proof the implementation is correct; process gates provide external evidence, and the human decision remains a separate third step. A quiz score never determines contract acceptance or a human-gate outcome.

What happens if a comprehension record was created without a real human answer?

Use `loopbench comprehension invalidate <task-id> --event <event-id> --by <name> --reason ...` to revoke that exact record while preserving the ledger history. Then repeat the applicable learning step; the original event is never edited or deleted.

When should the skill avoid asking the human a question?

Do not ask when the request, accepted contract, or standing instructions already decide, when testing can settle the question, or when answers would not change user-facing outcomes. File paths, test commands, and implementation methods are not human questions by themselves.