v0.14.0

Interview Mode for Human Gates

New mode: interview on human nodes enables structured multi-field form collection. An upstream agent generates questions with context and options; the interview handler presents them one at a time in a fullscreen form with progress tracking.

Highlights

• One question at a time — progress bar, answered summary, selection feedback (filled dot + checkmark)
• Structured JSON questions — agents output {"questions": [...]} validated before display, with markdown fallback
• response_format: json_object — force structured output at the LLM API level (Anthropic, OpenAI, Gemini)
• deep_review built-in workflow — 3 interview gates, parallel analysis (correctness/security/design), remediation plan
• Agent params map — generic key-value pass-through from .dip files (dippin-lang v0.16.0)
• Empty API response handling — session-level retry with diagnostics, hard-fail after 2 attempts

Interview UX

• Three field types: select (radio + Other), yes/no confirm, freeform textarea
• Enter confirms and advances, Ctrl+S submits all, Esc cancels with partial answers
• Text wraps to terminal width, pre-fill from previous answers on retry
• Cancel returns OutcomeFail so pipelines can route on cancellation

Fixes

• Autopilot hard-fails on provider errors (no more silent auto-approve)
• Goroutine leak in flashDecision fixed with done channel
• Mode 1 tea.Cmd propagation in all three TUI runners
• Context leak fix (explicit cancel() in retry loop)
• Empty response guard fails nodes instead of silent success

New Built-in Workflow: deep_review

tracker deep_review           # review any codebase
tracker --autopilot mid deep_review  # fully autonomous
tracker init deep_review      # copy and customize

Interview-driven codebase review: scope interview → parallel analysis → findings triage → priority interview → remediation plan.

Full changelog

See CHANGELOG.md for complete details.