Release v2.1.0 · BourbonDog/amicus

What's Changed

JSON output on resume/continue/abort: amicus resume, amicus continue, and amicus abort now support --json flag to emit structured, versioned documents matching the start --json schema. The abort command emits a new document type with scope and count information.
Did-you-mean suggestions: Typos in CLI commands now trigger helpful suggestions (e.g., amicus contnue suggests continue). Suggestions are capped at 3 and only shown for near-misses within edit-distance 2.

Agent polling guidance: All headless workflow documentation now recommends amicus_wait as the primary blocking call instead of sleep+status polling. The polling loop remains documented as an explicit fallback for clients without the wait tool.
MCP registration check fix: amicus doctor no longer false-negatives on healthy Claude Code registrations. The check now reads config sources directly instead of relying on discovery that strips its own entries.

Council review output sanitization: The synthesis leg now neutralizes model review text before passing it to the synthesis model, preventing instruction injection. Review output is wrapped in an explicit untrusted-data block.
Summary length optimization: Both council review waves now request --summary-length normal instead of verbose, reducing unnecessary token spend on paid API keys.
Workflow linting: Added actionlint (pinned to v1.7.7) to the CI quality job to lint GitHub Actions workflows and shellcheck all run: blocks automatically.

Corrected --agent defaults: Clarified that --agent defaults to Chat only in interactive mode, defaulting to Build in headless (--no-ui) runs.
Fixed council pipeline order: Updated commands/council.md to correctly reflect that council validate runs during Stage 1 (independent reviews) and council tally runs between Stage 2 and Stage 3.

Full Changelog: v2.0.0...v2.1.0