obra

Avoid repeating work by finding previous decisions, patterns, and solutions. Reduces context window usage by 50-100x through intelligent search of conversation history.

Build bulletproof AI skills using Test-Driven Development principles. Write tests first, watch them fail, then create skills that pass under pressure from subagents.

Engineers often struggle when joining a new codebase without context or clear implementation steps. This skill generates comprehensive implementation plans with bite-sized TDD tasks, exact file paths, complete code examples, and testing instructions for engineers with zero codebase context.

AI agents often claim work is complete without actual verification. This skill enforces evidence-based completion claims by requiring verification command output before any success assertions. It prevents false completion reports and builds trust through proof.

AI assistants often skip checking available skills before responding. This skill enforces a mandatory workflow where you must invoke relevant skills BEFORE taking any action, including asking clarifying questions. Prevents missed skills and ensures consistent, documented approaches are followed.

Develop multiple features in parallel without git branch switching. This skill creates clean, isolated worktrees with automatic project setup and test baseline verification.

AI assistants often skip tests and write code directly. This skill enforces test-first development by requiring failing tests before any implementation. It prevents bugs, documents behavior, and enables safe refactoring for production code.

Random fixes waste hours and create new bugs. This skill teaches a four-phase process that finds root causes before attempting fixes. It prevents rework by ensuring you understand the problem before proposing solutions.

Complex implementation plans are hard to execute reliably. This skill dispatches fresh subagents per task with automatic spec compliance and code quality reviews, ensuring nothing is missed and quality stays high.

Code issues found late in development are expensive to fix. This skill dispatches a code-reviewer subagent to catch problems early, before they cascade into larger issues.

Code reviews often contain unclear or technically questionable feedback that gets implemented without verification. This skill ensures AI evaluates feedback technically before implementation, preventing bugs and maintaining code quality.

Claude Code often repeats work or forgets solutions found in past conversations. This skill searches your conversation archive to retrieve past decisions, code patterns, and solutions without consuming your context window.

AI assistants often complete implementation tasks but lack structured guidance for integrating work back into the repository. This skill provides a step-by-step workflow for verifying tests, presenting merge and PR options, executing user choices, and safely cleaning up worktrees.

Complex multi-step implementations are challenging to manage safely. This skill breaks development plans into manageable batches with review checkpoints, ensuring quality and catching issues early.

Multiple independent failures waste time when investigated sequentially. This pattern dispatches one AI agent per independent problem domain, letting them work concurrently to solve separate issues in parallel.

Transform vague ideas into concrete implementation plans through structured dialogue. This skill guides you from concept to validated design with incremental validation at each step.