just-demand-verification

Verifies task outcomes against briefs and manages validation revisions before workflow closure.

Updated May 21, 2026
One-click install
npx skills add https://github.com/Sighthesia/just-demand --skill just-demand-verification-sighthesia
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: just-demand-verification
Source: https://github.com/Sighthesia/just-demand/tree/main/.opencode/skills/just-demand-verification
Command: npx skills add https://github.com/Sighthesia/just-demand --skill just-demand-verification-sighthesia

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Agents often claim tasks are done without fresh evidence, collapse verification failures into success, or guess at fixes when user feedback is vague. This Skill enforces an evidence-first verification gate so no workflow task closes without validated results. ## Core Features & Use Cases - Evidence-Gated Closure: Blocks completion claims without fresh verification evidence and routes failures to changes_requested, blocked, or debugging states instead of done. - Correction Feedback Handling: Translates vague user corrections into new validation revisions, routing ambiguous drift back to socratic-clarification rather than guessing at fixes. - Analysis Confidence Gate: Requires analytical conclusions to state confidence level, evidence basis, alternative explanations, and falsifiers to prevent false certainty. - Use Case: After an implementation pass, run this Skill to verify results against the task brief, produce a BLUF-style report under 300 characters, create a checkpoint commit, and archive the task via the complete-verification command. ## Quick Start Verify the current task implementation against its brief and report whether it is ready for closure.

Frequently Asked Questions about just-demand-verification

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

FAQPage Schema
How do I verify an AI agent's task before marking it done?▼

Run verification against the task brief and active validation revision, requiring fresh evidence such as test results or reviewer findings. The task only moves to done when verification passes; failures route to changes_requested, blocked, or debugging states.

What should I do when user feedback says the result feels wrong?▼

Treat vague correction feedback as a trigger for clarification, not a guessed fix. Create a new validation revision, present two-stage deviation options to locate the mismatch, and only re-implement after the expected effect and plan are confirmed.

How do I avoid overconfident analysis conclusions from an agent?▼

Require every analytical conclusion to state confidence level, supporting evidence, alternative explanations, and a falsifier. If evidence cannot distinguish between explanations, report insufficient evidence rather than a falsely definitive answer.

When should debugging lessons be captured as reusable skills?▼

Capture a lesson when an issue required three or more meaningful fix attempts or had a non-obvious root cause involving tools, state machines, caches, or concurrency. Skip capture for one-off business rules or module-only details.

Does verification automatically archive completed tasks?▼

Yes, the script-owned complete-verification command records the result, applies the checkpoint-commit safety gate, and archives the task package. Archival preserves the full package rather than destructively deleting it.