honey-stack v0.18.1

Fixed

• Keep learned guard warnings from becoming hard edit or command denials on Codex, and clarify that task inspection records do not restrict authorized edits.

honey-stack v0.18.0

Added

• Opt-in private Codex conversation capture with a per-user macOS observer, durable delivery records, explicit upload status and workspace-level disable controls.
• Capture works in plain workspaces as well as Git repositories and excludes pre-activation history, inherited subagents, injected instructions, reasoning, tools and repository files.

honey-stack v0.17.2

Fixed

• Keep UI-removal requests on ordinary verification while recognizing explicit application uninstalls.
• Reconcile published suggestion issues before automatic improvement work, preserving closure reasons without claiming the suggestion was implemented.

honey-stack v0.17.1

Fixed

• Prevent inferred inspection mode from blocking authorized edits or execution evidence.

honey-stack v0.17.0

Added

• Optional Delta history integration for Claude Code and Codex, with scoped search, bounded decision recall and fresh summary checks.
• Local handoff references that verify Git state and saved evidence, plus explicit capture-policy previews and separate upload readiness.
• Delta installation and usage guides, CLI regression tests and synthetic before/after QA.

honey-stack v0.16.4

Changed

• Merge pull request #29 from umararshad4/fix/intent-defaults-to-execute
• fix(tasks): honor negated edit verbs and keep read-only on task set
• fix(tasks): stop treating ordinary requests as read-only

honey-stack v0.16.3

Changed

• Merge pull request #28 from umararshad4/t3/docs-site-link
• docs: link the documentation site

honey-stack v0.16.2

Changed

• Merge pull request #27 from umararshad4/docs/compact-readme
• docs: compact README into a quick-start guide

honey-stack v0.16.1

Changed

• Merge pull request #25 from umararshad4/t3/fix-ship-github-issues
• fix(workflow): address gating and improvement backlog

honey-stack v0.16.0

Changed

• QA now defaults to the tester-army/e2e CLI and project MCP workflow, with complete upstream command references and focused, cache-aware reruns.
• Browser QA and verification discover WebMCP tools, with a same-session e2e adapter, isolated Chromium runner, explicit fallback and independent UI-proof requirements.
• QA replaces eligible setup with WebMCP, reuses verified sessions and tests, and limits separate contract checks to relevant work. Speed comparisons require matched workloads and repeated runs.
• QA can batch reviewed same-page WebMCP setup calls, builds complete case matrices before running, and avoids duplicate setup screenshots while retaining failure evidence.
• Fast QA uses native e2e reports, bounded isolated workers and deliberate tracing, and avoids polling values already read after confirmed readiness.
• WebMCP's browser launcher cancels its unused shutdown timer instead of adding two seconds of idle exit time. Fast QA distinguishes current-state assertions from negation stability windows and balances file/target units before adding workers.
• WebMCP QA runs on Playwright's headless Chromium build and automation switches, starts e2e while its browsers boot, gives each worker its own browser through an e2e browser provider, and enables Node's compile cache. The matched 24-case run is about 3.3x faster than the previous WebMCP flow.