Highlights

v0.6.0 is what the first real capture found. On 2026-08-20 fragcap ran against a Steam title for sixteen minutes and captured 18,234 packets, and that single run produced most of this release: the capture went silent for its whole duration, 91 percent of the file turned out to belong to an unrelated background process, and the summary reported none of it. Nearly every change below exists because something was observed to be wrong, not because it was imagined.

• Capture tells you what it is doing. A status block redraws in place at least once a second, showing elapsed time, the bound process, packets and bytes against any configured bound, the filter's narrowing state, every discard counter, and the top per-process contributors to the file so far. A redirected run gets a plain heartbeat instead of silence, and --json gains an optional capture.progress event. The sixteen-minute run whose last visible line was filter narrowed to 0 endpoint(s) is no longer possible.
• A capture contains the target's traffic. It did not reliably before. Output is now scoped to the target, what is excluded is counted rather than dropped quietly, and the completion summary reports what was written per process, so a capture dominated by something you did not ask for is visible in the file's own accounting rather than only under later analysis.
• Detection reports what it found, not what it guessed. Titles carrying no DRM stop being labelled "Steam DRM"; titles that genuinely ship Easy Anti-Cheat are now reported, which they were not, because the signatures only matched runtime files those titles never ship. Anti-cheat installed machine-wide as a Windows service is reported separately from any title's own evidence, never merged into it, since presence on the machine does not say which title put it there. A blank column now says which kind of blank it is, and a scan names what it did not cover. Nothing asserts a completed scan the tool cannot vouch for.
• A Steam title keeps all three of its names. The storefront name, the installed folder name, and the observed launch executable, none reconstructed from another, and any of the three resolves the title. On a sampled library 11 of 34 titles had a folder name that diverged from their storefront title. fragcap steam list also honors --json and prints column headers.
• --help stopped lying. Every flag that has a default states it, --sink names every scheme it actually accepts, the documented grammar matches the shipped one, and four new checks in the test suite keep it that way.

Breaking: fragcap catalog update is removed. It could not run in any released binary, and what it fetched was a third-party title list rather than any fragcap artifact. The three catalog seed verbs collapse into one catalog seed taking a repeatable --tier. Store paths are now overrides everywhere rather than requirements: seven subcommands that refused to run without an explicit path to a store fragcap already knows how to find now resolve it themselves, and every success line names the store it touched.