Kills self-repetition in LLM responses. Cuts 30-40% wasted tokens by eliminating restated questions, echoed instructions, recap paragraphs, and repeated ideas. Supports three intensity levels (lite, full, off) switchable via /parrot command. Auto-activates every session via SessionStart hook. Works with Claude Code's plugin system — install via claude plugin marketplace add animeshpatni94/parrot then claude plugin install parrot@parrot. Stacks with other skills like caveman for up to 80% token reduction.