Loopkit provides a structured set of skills designed to standardize AI coding workflows. It helps developers using agents like Claude Code, Cursor, and Gemini CLI maintain higher output quality through specialized passes for accessibility auditing and session memory management.
The toolkit focuses on the gaps in AI-generated code, offering specific routines to catch UI accessibility failures and a machine-readable decision tracking system to preserve context across sessions.
Follow the documented setup, then try a first task.
Installs: Complete Loopkit project bundle: the .claude harness, repository skills installed under .claude/skills, and available root harness files. Existing files are skipped by default. · Claude Code, Cursor, Codex, Gemini CLI
Run from: The project directory where Loopkit should be installed
In a terminal, change to the destination project directory, then run the installer. It installs into the current directory and skips existing files by default.
curl -fsSL https://raw.githubusercontent.com/Archive228/loopkit/main/install.sh | bash
Installs: Complete Loopkit project bundle, documented as installing the same files as the other project installation methods. · Any agent that reads SKILL.md
Run from: The current project directory
In a terminal, run the skills.sh installation command from the project that should receive Loopkit.
npx skills add Archive228/loopkit
Installs: The repository's .claude harness copied into a selected target project. This documented procedure does not copy the repository-level skills directory, so no skill-path mappings are asserted.
Run from: A parent directory from which both the new loopkit clone and target project directory are addressable
In a terminal, clone the Loopkit repository.
git clone https://github.com/Archive228/loopkit
After the clone completes, run this in the same terminal working directory. Replace the author-provided `your-project` placeholder with the path to the existing target project when entering the command; the documented command is preserved verbatim below.
cp -r loopkit/.claude your-project/
Uses adversarial-verify
Use the adversarial-verify skill to review the goal specification at [GOAL_SPEC_PATH] against the diff identified by [DIFF_OR_COMMIT_RANGE]. Do not run or change the code and do not propose fixes. Check every changed line for the documented shortcuts and return only the skill's JSON result.
Uses schema-diff
Use the schema-diff skill to compare [OLD_SCHEMA_PATH] with [NEW_SCHEMA_PATH]. Return a risk-ranked list of changes, explain data-loss, reversibility, and backfill concerns for destructive changes, and give a go/no-go recommendation with the safe rollout order.
Uses coverage-gaps
Use the coverage-gaps skill to inspect [CODE_SCOPE] together with [TEST_OR_COVERAGE_INPUT]. Identify the three highest-risk untested behavior paths, rank them by blast radius, and describe the test to add for each. Focus on error paths, boundaries, branches, and stateful ordering rather than the headline coverage percentage.
Claude
Codex
Cursor
Gemini CLI
Copilot
Windsurf