v1.7.0

Reliability release for framework-native terminal lifecycle behavior and reproducible evaluation evidence.

Highlights

Framework-native lifecycle boundaries

• Distinguishes final shutdown from resumable handoff for editors, shells, and foreground suspension.
• Adds current guidance for Bubble Tea 2.0.8, Ratatui 0.30.2, Textual 8.2.8, and Ink 7.1.1 across signals, child processes, re-entry, platform limits, full redraw, and external-state reload.
• Prevents terminal-reader races and preserves both child and restoration failures instead of allowing one outcome to mask the other.

Auditable evaluation provenance

• Introduces schema-v3 run, grade, and summary artifacts with stable JSON Schema identifiers, exact runner and grading provenance, privacy-preserving system-prompt metadata, and generation settings when exposed.
• Recomputes aggregate and per-case summaries from complete grades while retaining schema-v2 validation for historical v1.6.1 evidence.
• Documents the contributor evidence, privacy, review, and schema-evolution contract.

Lifecycle evaluation

• Adds five regression cases covering Bubble Tea editor handoff, Ratatui input ownership, portable Textual suspension, Ink terminal suspension, and cross-framework signal policy.
• A clean single-snapshot run with gpt-5.6-terra and high reasoning passed 25/25 human-graded assertions. This is conformance evidence for the recorded cases, not a statistical claim of general model lift.

Release verification

• Exact tag commit: 3d2a74630c42507123227e4df170fe4ba2abb841.
• The exact-tag checkout passed 18 harness/package tests, deterministic packaging, Skill Creator validation, ShellCheck, strict Claude plugin validation, and schema-v3 evidence validation.
• The installable archive contains the exact ten-file skill allowlist with normalized timestamps, modes, and order.

Asset SHA-256: f84abccf279fb54a9ef42acda1fbdec2b4b98ed8506ff4e94c700a3a41fa3e0b

See CHANGELOG.md for the full change list.

Full comparison: v1.6.1...v1.7.0