v0.6.10

What's Changed

• perf: stop doing work nobody reads (86ms to 39ms per command, 160MB of dead traces pruned) by @fajarhide in #285
• fix(distillers): route by what produced the output, not by the command's first word by @fajarhide in #286
• fix(cli): close the last six subcommands that swallowed unknown flags by @fajarhide in #287
• fix(store): name the gate that produced each passthrough row by @fajarhide in #289
• fix(jsts): keep the file:line eslint reported, not just how many there were by @fajarhide in #290
• fix(jsts): fold the bundler size table inside a declined composite buffer by @fajarhide in #292
• chore(repo): untrack the stray .snap.new and ignore future ones by @fajarhide in #294
• fix(mcp): stop offering one command's content as permanent noise by @fajarhide in #295
• perf(toml): compile a signal's line patterns only when that signal matches by @fajarhide in #296
• docs(metadata): stop selling the claim the README argues against by @fajarhide in #297
• chore: delete surface that nothing calls by @fajarhide in #298
• chore(deps): drop five dependencies nothing references (#174) by @fajarhide in #299
• fix(distillers): make the zero-state guard part of the type, and stop two distillers speaking for output they never read by @fajarhide in #302
• perf(hooks): stop loading a GPT vocabulary on every command to fill a reporting column by @fajarhide in #305
• fix(cloud): hand back kubectl listings, and stop counting OMNI's own banner as surviving content by @fajarhide in #306
• fix(jsts): stop sampling the head of output the distiller never read by @fajarhide in #311
• fix(paths, learn): stop the suite writing into the real ~/.omni, and a thread printing into a piped payload by @fajarhide in #308

Full Changelog: v0.6.9...v0.6.10