oberskills

CRITICAL - Invoke FIRST before ANY Task tool call. This skill gates all agent dispatch. If you are about to use Task tool, STOP and invoke this skill. Enforces oberprompt principles. Triggers on "Task tool", "launch agent", "spawn agent", "dispatch agent", "parallel agents", "subagent", "background agent", "Explore agent", "general-purpose agent", "Bash agent", "Plan agent".

Meta-skill for creating production-quality skills. Use when user wants to create a new skill, convert workflows into skills, or build multi-skill systems with routing. Triggers on "create skill", "build skill", "make me a skill", "skill for X", "convert to skill", "automate this workflow". Guides through checklist-driven creation with integrated pressure testing. Outputs single skills OR skill sets with router.

Use when encountering ANY bug, error, unexpected behavior, test failure, crash, wrong output, missing output, flaky behavior, race condition, regression, "it doesn't work", "something's wrong", "not working as expected", "should do X but does Y", broken functionality, logic error, runtime error, exception, assertion failure, timeout, hang, or any situation where code behavior differs from intent. Triggers on "debug", "fix", "broken", "failing", "investigate", "figure out why", "not working".

Execute approved plans using subagent orchestration. Use after oberplan produces a plan and user approves. Maintains a persistent execution checklist file that tracks all actions. Dispatches agents for each phase, enforces checkpoints with code-review subagents, and tracks progress. Triggers on "execute the plan", "run the plan", "implement the plan", "start execution", "dispatch agents for plan", or when you have an approved plan ready for implementation.

Quick hack mode for small features. Mini planning in-memory, then direct subagent execution. No plan files, no checklists, no checkpoints - just understand, dispatch, done. Use when user says "quick hack", "just build it", "hack this together", "prototype", "quick feature", or when scope is clearly small (1-2 files, single concern) and user wants speed over ceremony. NOT for multi-phase work, architectural decisions, or anything needing user confirmation.

Use when ANY planning, scoping, or design work is needed BEFORE implementation. Invoke when user asks to "build", "create", "implement", "add feature", "design", or describes desired functionality. CRITICAL - this skill dispatches FIRST, identifies relevant lens skills, clarifies requirements, and produces agent-executable plans. Triggers on "help me build", "I want to create", "let's implement", "design a", "plan for", "how should we approach", "figure out how to", or any request that requires understanding WHAT to build before building it.

MUST use before writing ANY prompt for Task tool, subagents, or agent dispatch. Use when writing prompts, system messages, agent instructions, hooks, or skills. Use when prompts produce inconsistent results, wrong outputs, or hallucinations. Use when optimizing prompts for accuracy, calibration, or efficiency. Use when debugging broken prompts. Triggers on "improve my prompt", "prompt engineering", "system prompt design", "agent instructions", "Task tool prompt", "subagent prompt", "agent dispatch", "flaky LLM outputs", "overconfident responses", "prompt not working", "hallucinating", "prompt injection defense".

Multi-dimensional web search orchestrator. Use when the user needs comprehensive web research that would benefit from parallel searches across different angles. Dispatches multiple fast subagents (haiku) to search in parallel, synthesizes results, and returns only relevant information with source URLs. Preserves main agent context by handling search complexity in subagents. Triggers on "research this", "find information about", "search the web for", "comprehensive search", or when a single search wouldn't capture all relevant aspects.