agent-superpowers_skill

Overview

This skill transforms a general-purpose OpenClaw agent into a disciplined engineering assistant. It enforces verification before claims, a three-agent review pipeline, systematic debugging, brainstorming gates, and anti-over-engineering constraints to keep work focused and reversible. Use it to raise reliability, reduce wasted effort, and make agent-driven development auditable.

How this skill works

The skill injects behavioral rules that require evidence for any completion claim and a stepwise pipeline: brainstorming → plan → implement → spec-review → quality-review → verify → complete. For implementation tasks it spawns three sub-agents (implementer, spec reviewer, quality reviewer) and enforces a four-phase debugging flow with a three-strike stop rule. Verification gates require running explicit commands and including command output before declaring success.

When to use it

• Implementing coding tasks of any complexity where verification and reviews matter.
• Debugging failures, test regressions, or hard-to-reproduce bugs under pressure.
• Before claiming work is complete, to produce independent evidence of success.
• When spawning sub-agents to execute plan steps and maintain auditability.
• Planning features or components to force a short, approved design before coding.
• Performing code reviews or enforcing codebase patterns and quality gates.

Best practices

• Always present a short design and get approval before implementing any creative work.
• Write minute-granular plans (2–5 minute steps) with exact file paths and commands.
• Dispatch three reviewers in sequence: implementer → spec reviewer → quality reviewer.
• Reproduce errors, investigate root cause, test one hypothesis at a time, and stop after three failed fixes.
• Run the proving command and paste full output before stating tests/builds/bugs are fixed.
• Prefer local, reversible changes; confirm before hard-to-reverse or external-facing actions.

Example use cases

• Add a new feature: propose 2–3 approaches, get design approval, then follow the plan and review pipeline.
• Fix a failing test: reproduce, instrument boundaries, hypothesize, change one variable, and verify with test output.
• Spawn implementer/spec/quality sub-agents to complete and independently verify a complex refactor or feature.
• Claim completion of a ticket only after running the defined verification command and including its output.
• Avoid scope creep by following anti-over-engineering rules when asked for small, targeted fixes.