croftspan

Deep code sweep — dispatches 3 parallel focused auditors for security, stubs, and code quality. Works standalone or offered after gigo:execute completes. Use gigo:sweep.

Autonomous UI polish loop — screenshots page at multiple viewports, compares against visual spec, fixes issues, re-screenshots until criteria met or 5 iterations. Internal tooling for GIGO's own site and docs.

Execute implementation plans. Lead dispatches bare worker subagents per task, invokes gigo:verify after each, tracks progress via checkpoints. Falls back to inline if subagents unavailable. Agent teams available as experimental opt-in. Use when you have an approved plan from gigo:blueprint.

Test whether your assembled context actually improves AI output. Runs tasks bare vs assembled, uses comparative judging, reports the delta. Use gigo:eval when you want to prove your setup works or debug why it doesn't.

Brainstorm and produce approved design briefs. Use when the user has an idea, feature, or task that needs planning — from vague intent to approved design. Hands off to gigo:spec for formal spec and plan writing. Use gigo:blueprint.

Ongoing maintenance for your assembled expert team. Add expertise, audit for bloat, restructure messy setups, or upgrade older projects. Auto-detects severity — targeted addition, health check, or full restructure. Use gigo:maintain, /maintain, or when gigo:blueprint or gigo:snap detect gaps.

Session retrospective — reads Claude Code telemetry and turns friction into project improvements. Analyzes one session or aggregates recent sessions. Produces proposals, not reports. Use gigo:retro.

Assembles an expert team and scaffolds a Claude Code project for any domain — software, writing, game dev, research, design, anything. Researches the best practitioners in the field, blends their philosophies into focused personas, and writes lean .claude/ project structure with rules, standards, and workflow. Use this skill when the user wants to start a new project, set up Claude Code for a proj

Run The Snap — audit rules for bloat, staleness, and derivability, then capture session learnings. Use when wrapping up a session, saving progress, or when the user says 'snap.' Requires an existing .claude/rules/snap.md in the project.

Formalize an approved design brief into a spec and implementation plan. Self-reviews both, runs Challenger for large tasks. Use after gigo:blueprint produces an approved brief, or when you have your own design brief to formalize. Use gigo:spec.

Two-stage review: spec compliance (did you build the right thing?) then craft quality (is the work well-built?). Invoked automatically by gigo:execute after each task, or standalone on any work. Use gigo:verify.