proactive-task-validator

Validating proactive assistant tasks for checklist leakage requires checking whether Step 1 and Step 2 satisfy checklist items using hidden and dependency intents, while Step 3 prompt-only outputs fail. This Skill orchestrates isolated candidate generation and centralized judging to verify this three-step logic.

Dependency-group isolation works by building per-task validation packets that prevent checklist leakage across tasks. It isolates candidate outputs via subagents for each dependency group before centrally judging Pass/Fail/N/A verdicts with itemized checklist coverage.

Generating an audit report for long-horizon proactive workflows involves running a two-stage validation process on episode.yaml datasets. It produces a reviewed Markdown report containing structural warnings, per-task verdicts, and repair guidance for weak tasks.

Yes, you need Python3 and PyYAML to run dataset validation on episode.yaml files. The Skill also depends on standard libraries including json, pathlib, and argparse to parse task structures and execute validation scripts.

This validation approach is specifically designed for datasets structured as episode.yaml with proactive and dependency-heavy evaluations. Datasets without dependency intents or hidden context lack the required structure to generate per-task dependency-group validation packets.

Prompt-only candidate generation fails during checklist validation because the logic intentionally verifies that checklist items cannot be satisfied without hidden intent and dependency intent context. This ensures prompt-only information is insufficient to pass the evaluation.