inspect

Reviews build plan documents adversarially through fresh inspector models before construction begins.

Updated Jun 2, 2026
One-click install
npx skills add https://github.com/line7works/tony-skills --skill inspect-line7works
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: inspect
Source: https://github.com/line7works/tony-skills/tree/main/plugins/inspect/skills/inspect
Command: npx skills add https://github.com/line7works/tony-skills --skill inspect-line7works

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes assets (resource) components.

What problem does it solve? A flawed build plan approved without review is the most expensive defect in a spec-driven development loop, because the builder faithfully implements whatever the plan says. This Skill catches plan defects — invented requirements, uncheckable criteria, wrong repo facts — before any code is written. ## Core Features & Use Cases - Adversarial plan review: Sends the build doc and scope doc verbatim to fresh inspector models (Claude, GPT, Gemini, DeepSeek, or Qwen) that hunt for reasons to reject the plan across three lenses: traceability, code-book compliance, and repo reality. - Verified findings: Every BLOCKER and MAJOR finding is verified against cited line numbers before reaching the user, with refuted findings counted and dropped. - Paper trail: Appends a punch-list block and a model-named inspection stamp to the build doc, keeping a full audit history without ever editing the plan itself. - Use Case: After drafting a build plan with /blueprint, run /inspect to have an independent GPT or Claude inspector grade the plan against the blueprint rules and the actual repository before /build executes it. ## Quick Start Type /inspect with the path to a build plan document to run an adversarial pre-construction review of that plan.

Frequently Asked Questions about inspect

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

FAQPage Schema
How do I review a build plan before implementation?

Invoke /inspect with the path to a build plan document. The skill sends the plan and its scope doc verbatim to a fresh inspector model you choose, which hunts for traceability gaps, rule violations, and incorrect repository claims, then reports a verified verdict.

What AI models can review my plan document?

The inspector roster includes Claude, GPT, Gemini, DeepSeek, and Qwen, selected per run through a numbered prompt. Claude runs three lenses as one fleet; outside models cover the document lenses while a local Claude call checks repository reality.

Does the inspect skill modify my build plan?

No. It makes exactly two additive writes: a punch-list block of findings and an inspection stamp naming the model and severity counts. It never edits plan content, fixes issues, or changes Status lines — the plan author adjudicates and amends.

What happens if there is no scope document for the plan?

The inspection still runs under a no-record rule. Untraceable requirements become Questions asking for confirmation rather than automatic BLOCKERs, since absence of a record is not proof the detail was invented, and the verdict notes the run is weaker.

Why does inspect require the readers plugin?

All inspector calls are dispatched through the readers plugin's roster, which owns model ids, transports, sandboxes, and authorization. The inspect skill never sends model requests directly, so readers must be installed for any lane to run.