fix: remove broken tr pipe that corrupts UTF-8 in bounty responses

fix: aggressively prevent match-joining when user requested bounties

fix: use macOS-compatible tr for control char stripping, strengthen bounty polling rule

fix: add revert handling and jq workaround to SKILL.md

fix: replace broken cast call with API lookup for agent ID

fix: clarify bounties are free to join in skill docs

fix: strip all control chars including newlines from API responses

fix: restructure section 3.6 to prioritize user intent over auto-detect

fix: bounty mode — zsh quoting, respect user intent, claim reward

fix: sanitize curl output before jq to prevent control char parse failures

deploy: update SKILL.md with BountyArena mainnet address

fix: SKILL.md v3 refinements for bounty instructions

feat: SKILL.md v3.0.0 — bounty loop, budget strategy, reputation

Align README prompt with frontend CTA, remove dual install flow

Fix 4 AI-confusion points in SKILL.md