Skip to content

KonX 0.27.0

Choose a tag to compare

@256kMagic 256kMagic released this 23 Oct 20:29

KonX 0.27.0 — Reviewed Copilot Actions

2026-10-23. KonX adds reviewed diagnostics, undoable editor replacements and saved answer notes to Copilot.

Changes

  • Review and run a bounded single-target Ping, DNS query or Traceroute from a completed Copilot answer.
  • Review an exact editor selection replacement, then recheck the live model, version and selection before one undoable edit; no automatic save or upload.
  • Save an exact answer or selected excerpt as a new local Knowledge note, with provenance.
  • Retain action targets, review digests and outcomes. Expired, stale and replayed reviews are rejected; uncertain outcomes never trigger an automatic retry.
  • Avoid the reproduced Windows initial-publication failure in Copilot, Incidents, Change Validation and saved sites with a native move that refuses to replace an existing destination. Preserve source identity checks and ownership after failed cleanup.

Read the Copilot guide and user guide.

Downloads and upgrading

Windows x64: installer, portable, and checksums.

Installed 0.6.2+ and portable 0.6.0+ can use Update KonX. Installed 0.6.0/0.6.1 requires one manual upgrade. Save drafts, stop listeners, and disconnect before restarting; the existing profile remains in use.

Validation

Full suite: 1,063/1,063 passed. Windows packaging verified 157 source files and 9 runtime dependencies. Packaged SSH passed 28 negotiation and 16 key-format checks; serial passed 27 synthetic checks. Unpacked and portable builds passed isolated save/restart smoke checks. Executable product versions match 0.27.0.

Focused checks exercised real loopback Ping, DNS and Traceroute, actual Monaco edits and one-step undo, local Knowledge persistence, stale/replayed reviews, cancellation, lost acknowledgments, shutdown and pending-action recovery. Model responses were synthetic; no paid provider request was made.

Both packaged builds executed a reviewed loopback Ping, saved the exact reviewed note, applied and undid the selected editor change, then restarted and retained all three outcomes without repeating them. The smoke harness initially placed its restart assertion in the save phase and omitted an editor dropdown change event; those fixture errors were corrected, their logs retained, and the complete checks rerun.

The first full run passed 1,061/1,062 tests. Improved failure logging then reproduced the remaining saved-site failure: the first update after importing a workspace hit the same Windows publication error. The draft was correctly retained, which prevented the next delete confirmation expected by the test. The shared publication fix was extended to saved sites before rebuilding and repeating the release gates. Full errors and fixture failure artifacts are now retained instead of being hidden by a four-line log limit.

Coverage

Automated checks use isolated profiles, synthetic provider credentials and responses, offline data, and owned loopback peers. Live-provider answer quality, physical equipment, and an elevated installed upgrade remain unverified. Executables remain unsigned. Answers display after complete redaction, even for streaming transports. Redaction recognizes configured credentials and known formats; arbitrary sensitive text still requires review.

An isolated Electron stress probe reproduced 30 failures in 500 replacements after initial hard-link publication and removal of the temporary alias, with tracked handles closed. The native no-replace move control completed 500/500 replacements. With the corrected publication boundary, the real action-store workload completed 500 lifecycles and 2,000 replacements without failure. Original failure evidence was retained. Native regressions also cover existing destinations, long Unicode paths, changed source versions, permanent failure and retained cleanup. These measurements apply to this Windows host and Electron runtime; persistent external file locks can still produce a visible save error.

Terminal command execution remains outside the action catalog. Commands in answers remain text for manual review; the existing session identity does not establish that the remote target and command mode are still appropriate. Editor actions modify only the reviewed live selection and never save or upload a file automatically.