v0.8.0

0.8.0 (2026-10-01)

Features

• add jev_ask, the caller-written ask extension tool (ADR-0077) (ea475dd)
• add jev_file_judge, the server-read file judgment tool (ADR-0077) (4a5df4c)
• add jev_files_judge, the batched file judgment tool (ADR-0077) (5d19992)
• add the opt-in hook screen PostToolUse annotator (ffb2d90)
• cut the compaction summary at the live-work turn (ADR-0077) (6adb044)
• route hook gate through guard questions (99c3571)

Bug Fixes

• deny pattern-matched write content before any provider call (d2d5cca)
• judge only real user turns over a redacted transcript tail (6b3080d)
• keep credential redaction linear on long identifier runs (0df5b30)
• make jev_ask command execution opt-in, denylisted, scrubbed, and bounded (53eac60)
• reconcile jev_files_judge with the landed jev_ask (ADR-0077) (7e0486b)
• refuse every *.env dotenv as a secret store (e9be1e6)
• refuse secret stores and redact credential literals in file state (09e7f9e)
• refuse to work when a moved clone leaves core.hooksPath stale (794024b)
• scan write content with a strict credential-literal detector (5f9392c)
• stop at the caps and summarize the overflow (ADR-0077) (04e9c7f)
• target screen at injected instructions, not any instruction text (524cb3f)
• tighten the credential-literal detector against probed false positives (3dd75c0)

Documentation

• add composite, routing, and catalog-traversal recipes (774b7ed)
• correct the dogfood confinement record (f70e7cc)
• pass the S8 dogfood gate and scope the ask-command denylist (efd933b)
• record the ask and file-judgment extension family (ADR-0077) (afb6fed)
• record the hook guard-question design in ADR-0076 (d81d937)