Wield Overview
A comprehensive engineering toolbelt for Codex and Claude Code — 26 reusable skills, Claude Code agents and workflows, and unobtrusive git hooks for the full software lifecycle.
Wield
Wield is a comprehensive engineering toolbelt for Codex and Claude Code. It
bundles the full software lifecycle — planning, building, reviewing, debugging,
shipping, researching, and documenting — into 26 reusable skills. Claude
Code also receives 13 specialized agents and 2 dynamic workflows; Codex
uses native subagents and equivalent portable skills for delegated work. Both
hosts share the same hooks and project rules.
Invoke skills as /wield:<name> in Claude Code or $wield:<name> in Codex.
Wield v2 supports Codex and Claude Code. Nothing is required to get value on day one — every skill has sensible defaults. See the host-specific installation instructions in Getting Started.
What you get
26 skills
Reusable workflows for planning, building, reviewing, debugging, researching, documenting, designing, testing, and deploying.
13 Claude Code agents
Focused named subagents for planning, implementation, review, testing, debugging, research, docs, and git. Codex uses native subagents.
2 multi-agent workflows
Claude Code scripts for /ultra-review and /sweep; Codex provides the same behaviors as $wield skills.
Guard hooks
Privacy and context-scout hooks plus session and uncommitted-change reminders — all fail-open and silent when Node isn't installed.
Shared rules
Engineering conventions several skills and agents reference at runtime: development rules, workflow, orchestration, and skill routing.
The idea
A single general assistant doing every phase of engineering — planning, implementation, review, testing, shipping — tends to lose the plot on anything larger than a quick edit. Wield splits that work into focused skills (what to do) and focused agents (who does it), so each phase gets a tool built for it.
You start with a broad prompt, jump straight to code, and rely on memory for what still needs review, testing, or documentation. Fast for tiny tasks, harder to trust as the work grows.
You route work through explicit skills (/wield:implementation-planning,
/wield:feature-building, /wield:code-audit, /wield:release-pipeline; use
$wield: instead in Codex), backed by specialized agents and shared rules.
Slower for trivial edits, more
reliable for real engineering.
How the pieces fit
- Skills describe a reusable workflow:
/wield:<name>in Claude Code and$wield:<name>in Codex. - Agents are the named subagents Claude Code dispatches for focused work; Codex uses its native subagents for the same delegated work.
- Workflows are Claude Code scripts that fan out across many agents and verify
their own findings — installed separately and invoked as
/<name>. Codex ships their equivalents as$wield:ultra-reviewand$wield:sweepskills. - Hooks guard your context and secrets, and nudge you about git state.
- Rules are shared conventions skills and agents load at runtime via
${CLAUDE_PLUGIN_ROOT}/rules/.
Where to start
Get Started
Install Wield and run your first workflow.
Browse the Skills
All 26 skills, grouped by what you're trying to do.
Meet the Agents
The 13 Claude Code subagents and when Claude reaches for each.
Run the Workflows
Claude Code workflows and their Codex skill equivalents.
Tips & Tricks
Patterns for chaining skills, using flags, parallelizing safely, and tuning config.
System Reference
Hooks, rules, and configuration internals.
License
MIT © Denis Ibraliu