Skills Reference

Skills are slash commands that trigger specialized behavior in Claude Code. Type a skill command in the Claude Code prompt and it runs a structured procedure designed for that task.

ForgeKit skills#

Command	What it does
/build	Compile the project via UBT and parse errors with UE-specific categorization
/map-codebase	Scan Source/ for UCLASS/USTRUCT/UENUM declarations and populate the Dev wiki
/map-design	Discover characters, abilities, and weapons from source code and content folders
/state-save	Capture full session state (VCS, build, tasks, decisions) before clearing context
/state-load	Restore saved session context and detect VCS drift since the last save
/review	Review changed files using a game development quality rubric (CRITICAL / WARNING / SUGGESTION)
/validate	Self-validating loop: spawns parallel agents to verify a plan, spec, or analysis
/health-check	Validate environment integrity across 18 structural and 12 semantic checks
/update	Analyze self-learning friction logs and propose environment improvements
/setup	First-time project configuration wizard: detects engine path, VCS, GAS, and plugins
/remember	Store a quick note in working memory with optional category tagging
/recall	Search across Design wiki, Dev wiki, working notes, and decisions

Bundled skills#

These three skills are included with Claude Code and available out of the box. They are not ForgeKit-specific: no installation is required and they work in any Claude Code project.

Command	What it does
/simplify	Review and simplify recently changed code for clarity and maintainability
/batch	Run multiple operations or commands in sequence
/debug	Diagnostic workflow for investigating and resolving build or runtime errors

How skills work#

Skills are defined in .claude/skills/<name>/SKILL.md. When you type a skill command, Claude Code reads the corresponding SKILL.md file and executes the procedure described in it. Each skill specifies:

• Trigger phrases: the exact commands and natural language phrases that activate it (e.g., /build, "compile", "check if it compiles")
• Negative triggers: phrases that look similar but should NOT activate the skill (e.g., /build should not trigger for "build documentation")
• Allowed tools: which Claude Code tools the skill can use
• Context mode: fork means the skill runs in a subagent with its own context window, keeping the main conversation clean

Skills run as subagents (context: fork). This means each skill invocation gets a fresh context window, so long skill procedures like /map-codebase do not consume your main conversation context.

Skills vs agents#

Skills are commands you invoke. Agents are specialists the orchestrator routes to. When you say /build, you are invoking a skill directly. When you say "this ability is crashing", the orchestrator decides to delegate to the debugger agent. Both run as subagents under the hood.