v0.12.2
Fixed
--openno longer fails silently on WSL. Spawning the OS
opener withstdio: "ignore"+ immediateprocess.exit()meant
the user never saw the spawn error whenxdg-openwas missing
(typical on a headless WSL) or exited non-zero.openInDefaultApp
now (1) sync-checks the command is on PATH, (2) preferswslview
when WSL is detected and the binary is installed, (3) waits up to
200ms for the spawned child to fail fast and surfaces botherror
events and non-zero exit codes on stderr.summaryon a day with zero activity no longer spends LLM
tokens.report --date todayalready returns "No activity
found";summarywas happily piping that empty payload to claude
for a useless answer (and-oopened an empty page). The daily
path now matches the range path's "no activity → skip" behavior:
writes a stub to the cache, prints a clear stderr line, returns
success without spawning claude.
Added
Smoke (Windows)workflow. Manual-trigger CI job that
exercises the real CLI on a Windows runner (agenthud --version / --help / --once / --cwd / report / summaryagainst a faked
session and a pre-seeded cache file). The unit-test matrix already
covered Windows, but it can't catch path-separator regressions in
stderr labels,cmd /c startquoting, or the filesystem side of
regenerateIndex— this one can. Trigger from the Actions tab.- README "Platform notes" paragraph clarifying that macOS and
Linux are the daily-driver targets, Windows is best on WSL2
(Anthropic's recommendation for Claude Code itself), and native
PowerShell may need an ExecutionPolicy adjustment for any
npm-installed CLI.
Fixed
- Windows:
prompt = ...stderr line used backslash separators.
formatPromptSourcenow normalises its output to forward slashes
so the user-facing label reads as~/.agenthud/summary-prompt.md
on every platform. (Pure cosmetic on POSIX; lets the Windows
CI tests pass.)
v0.12.0 was tagged but never reached npm — the Windows test job in
CI caught this issue before the publish step ran. v0.12.1 is the
first 0.12 release on npm.