Gap Map v0.1.7
Gap Map v0.1.7
Download → gapmap.myind.ai
Installers:
- macOS Apple Silicon —
Gap-Map-*-macOS-Apple-Silicon.dmg - macOS Intel —
Gap-Map-*-macOS-Intel.dmg - Windows —
Gap-Map-*-Windows.msi(managed deploy) orGap-Map-*-Windows-Installer.exe(click-to-install) - Linux —
Gap-Map-*-Linux.AppImage(portable) orGap-Map-*-Linux.deb(Debian/Ubuntu APT)
What's changed
New
- User-driven topic merge — merge two topics (and their posts, findings,
and chats) from the GUI or CLI, with a dry-run preview of exactly what moves
and a clear "this cannot be undone" confirmation. - MCP "Copy config" button (Settings) — copies the exact
mcpServersentry
Connect would write, so you can paste it into any MCP client by hand. Backed
by a newmcp_config_snippetcommand that mirrors the installer's path
resolution and refuses to hand out an unstable/translocated path. - MCP-disconnect diagnosis — clearer guidance when the app runs from a DMG
mount or a macOS App-Translocation path (move to /Applications and reopen).
Fixed
- WAL data-loss fix (critical) — the SQLite WAL self-heal no longer deletes
the live-wal/-shmside files; deleting them while another process held
the database open could silently discard committed transactions. (This had
cost ~56ktopic_postsrows in testing; recovered.) - Security hardening —
open_urlnow allows onlyhttp(s)/mailto
schemes; CSPscript-srctightened; markdown rendering XSS closed; user
values bound as SQL parameters (no string concatenation). - Worker robustness — an out-of-memory exit (137) is now treated as
recoverable and no longer counts toward the give-up threshold. - Build fix — created the missing shared
src/lib/toast.jsmodule that the
topic-merge modal imports (was breaking the production frontend build). - Test isolation — the topic-merge test fixture now uses an isolated
temp database (GAPMAP_DATA_DIR+tmp_path) and can no longer mutate the
real application database.
CI / Release
- Race-proof public-release promotion: a dedicated
workflow_runjob now flips
the public release draft → latest once mac + windows artifacts are present,
instead of relying on per-job timing.