An agent harness that turns you into a manager of autonomous digital engineers.
Skill Claude CodeCodex
Collaborate with Claude (Anthropic Claude Code model) for cross-model review and parallel problem-solving. Supports two modes - review (Claude reviews your work) and parallel (both agents work independently then compare), with multi-round continuity by recording Claude's sessionid and resuming it.
Skill Claude CodeCodex
Multi-agent code review aligned with ultrareview. Dispatches five specialist reviewers in parallel (perf, robustness, standards, spec, proof-obligations), then runs a per-finding validator pass that independently reproduces every finding before it reaches the report. Use when asked to perform a code review.
Skill Claude CodeCodex
Collaborate with Codex (OpenAI GPT model) for cross-model review and parallel problem-solving. Supports two modes — review (Codex reviews your work) and parallel (both agents work independently then compare).
Skill Claude CodeCodex
Use this skill BEFORE starting any software engineering task that will span more than one step. It creates .agent/ /plan.md with goal, success criteria, and stepwise plan; presents the plan for approval; and drives execution step by step. Invoke when starting or resuming a new feature, investigative bug fix…
Skill Claude CodeCodex
Process review comments on an MR, a design doc, or a spec — validate each finding independently with parallel validator subagents, apply fixes for the ones that hold up, and produce a reply report designed to earn reviewer sign-off. The report explains exactly how each confirmed issue was fixed, and for anything not…
Skill Claude CodeCodex
Capture lessons that would make a similar future task faster, more accurate, or less error-prone, and suggest where each lesson should live (rule, skill, or memory) for human approval. Use proactively when you catch yourself making a wrong assumption, get corrected, hit a failure, or notice something non-obvious that…
Skill Claude CodeCodex
Creates and fills spec.md sections 1-3 (background, goals, requirements) for new features and significant changes. Invoke when the task warrants a design doc — typically new features, significant refactors, or cross-module changes. Also triggered when the user explicitly asks to write a spec or clarify requirements.
Skill Claude CodeCodex
Fills spec.md sections 4+ (design, alternatives, test plan, observability). Invoke after sections 1-3 are approved at Gate: Requirements Understanding. The agent researches the codebase, drafts the design, and presents at Gate: Solution Design for approval.
Skill Claude CodeCodex
Breaks a spec.md into an implementable task list (tasks.md) with dependencies, context, and acceptance criteria. Invoke after the design is approved at Gate: Solution Design, before starting implementation.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: