Review plans, code, and workspace states to identify defects and propose fixes.
npx skills add https://github.com/mishahanin/heading-os --skill scrutinize-mishahanin
Please help me install this Agent Skill. Skill: scrutinize Source: https://github.com/mishahanin/heading-os/tree/main/.claude/skills/scrutinize Command: npx skills add https://github.com/mishahanin/heading-os --skill scrutinize-mishahanin
This Skill provides a rigorous, adversarial review gate for plans, code, and workspace states, preventing high-stakes errors and drift that standard agentic workflows often miss.
Invoke the scrutinize skill to perform a principal-engineer review on the current workspace to identify potential risks and structural improvements.
High-intent search queries and answers about installing and using this skill.
Adversarial code review uses cross-family judge rotation to debate findings and filter false positives, ensuring high-confidence defect identification. This process blocks forward progress until all high-severity issues are addressed and approved.
A principal-engineer review gate is a maximum-effort validation pass that stress-tests architectural plans to identify structural risks and drift. It executes a Validate-Identify-Improve-Adjust sequence to ensure structural integrity and propose evidence-backed fixes.
Yes, workspace validation requires Python3 and Git as dependencies, along with Langfuse for tracing. This environment setup provides the necessary execution and version control context to perform maximum-effort code and plan reviews.
Adversarial refutation works by using cross-family judge rotation, specifically Claude and Kimi k3, to debate high-severity findings. This mechanism filters out false positives by requiring multiple models to refute or validate each identified defect.
Yes, the review gate explicitly blocks forward progress until all high-severity issues are addressed and approved by the user. This prevents high-stakes errors from advancing through standard agentic workflows into production.
Use adversarial testing before approving complex architectural plans or deploying major refactors, as standard agentic workflows often miss drift and high-stakes errors. This approach catches potential regressions by stress-testing logic through maximum-effort validation.
chaimtInitialize a Git repository with an initial commit in the current directory.
greglambDispatch a code-reviewer subagent to identify defects and readiness issues in completed code changes.
yocxy2Automate testing, adversarial review, changelog generation, and pull request creation.
apexradiusVerify deliverable done-claims with evidence-backed adversarial checks.
