ironclad

Reviews GitHub plan issues and pull requests with a panel of parallel subagent lenses.

Updated May 3, 2026
One-click install
npx skills add https://github.com/Claudfather/clauDNA --skill ironclad-claudfather
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ironclad
Source: https://github.com/Claudfather/clauDNA/tree/main/skills/ironclad
Command: npx skills add https://github.com/Claudfather/clauDNA --skill ironclad-claudfather

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Plans and pull requests often ship with hidden blockers, scope drift, duplicated abstractions, or ignored prior art because a single reviewer cannot apply every critical lens at once. This Skill dispatches a panel of independent review lenses as parallel subagents against a GitHub Issue or PR, aggregates their findings, and posts a single deduplicated review comment with a convergence verdict. ## Core Features & Use Cases - Multi-lens panel review: Dispatches lenses such as first-principles, align-to-mission, extension-check, precedent-check, plan-health-audit, and cost-benefit as parallel subagents, each emitting structured findings. - Hardening loop for plans: With --loops N on a plan Issue, it repeats dispatch, folds findings back via forge --reforge, and re-checks convergence until no blockers and all decision forks are locked. - Single-lens and automation modes: --lens runs one lens for a focused report without posting a comment, and --auto emits a machine-parseable JSON structured-result block for CI or bot callers. - Use Case: Point it at a freshly drafted plan Issue with /claudna:ironclad <issue-url> --loops 3 to have the panel challenge the plan, re-forge it, and confirm convergence before implementation begins. ## Quick Start Ask the agent to run the ironclad review on a GitHub issue or PR URL, for example by saying: run ironclad on https://github.com/owner/repo/issues/42 with two hardening loops.

Frequently Asked Questions about ironclad

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

FAQPage Schema
How do I review a GitHub pull request with multiple review lenses?

Run the skill with the PR URL, for example /claudna:ironclad <pr-url>. It classifies the target, dispatches applicable lenses as parallel subagents, aggregates their findings, and posts one deduplicated review comment on the PR.

How do I harden a plan document before implementation?

Run the skill against the plan Issue URL with --loops N, such as --loops 3. Each cycle dispatches the lens panel, folds findings into the plan body via forge --reforge, and re-checks convergence until no blockers remain and all decision forks are locked.

What do I need installed to run this GitHub review workflow?

You need the GitHub CLI (gh) installed and authenticated. The skill runs gh auth status as a pre-flight check and stops immediately with an error if the CLI is not logged in, since every phase reads or writes GitHub through it.

Can I run just one review lens instead of the full panel?

Yes, use --lens <name> with a lens from the panel table, such as --lens cost-benefit. It dispatches only that lens and presents the structured report in the session without posting a comment or running a convergence check.

What happens when a review lens fails during a run?

Each failed lens is retried once as a fresh subagent. If the retry also fails, the run proceeds with partial results, and a single lens failure does not block the aggregated comment or the convergence decision.

Does this skill edit the plan or PR code directly?

No. It never authors the plan body or edits PR code; it only posts comments and orchestrates subagents. Plan body rewrites are delegated to a dispatched forge --reforge subagent, and code changes are left to humans or the build workflow.