Skip to content

Releases: treadiehq/gpu-kill

v0.1.20

04 Mar 15:40

Choose a tag to compare

What's Changed

Bug Fixes

  • fix(#47): Rogue detection now fully respects configured enabled_detections flags, scoring weights, risk_thresholds, and custom_patterns — previously these config fields were silently ignored and hardcoded defaults were used instead
  • fix(#48): SSH debug logging no longer leaks plaintext passwords when using sshpass; stderr from successful SSH commands is now surfaced as a warning instead of being silently discarded
  • fix(#48): SSH streaming (--watch mode) simplified from manual thread-based io::copy to Stdio::inherit() — fewer threads, no buffering overhead
  • fix(#49): MCP JSON-RPC requests with "id": null are now correctly rejected with an error instead of being silently treated as notifications (causing clients to hang waiting for a response that never arrives)
  • fix(#49): ToolContent and ResourceContents optional fields no longer serialize as "field": null, fixing compatibility with strict MCP clients

Internal / CI

  • fix(clippy): Replace &[snapshot.clone()] with std::slice::from_ref(&snapshot) in coordinator
  • fix(clippy): Gate Instant import behind #[cfg(unix)] in proc.rs to fix Windows unused-import warning
  • fix(clippy): Add #[allow(dead_code)] to detect_rogue_activity_from_records (public API reserved for future cluster-wide rogue detection)

v0.1.19

18 Feb 23:27

Choose a tag to compare

Release v0.1.19

v0.1.18

18 Feb 23:09

Choose a tag to compare

chore: release v0.1.18

v0.1.15

04 Feb 03:04

Choose a tag to compare

Summary

  • fix CI formatting failure by applying rustfmt to rogue detection changes

Tests

  • cargo build
  • cargo test

v0.1.14

04 Feb 02:49

Choose a tag to compare

Summary

  • fix CI build errors by using NVML helper functions for top process selection
  • make MCP GPU process type clonable for resources list aggregation

Tests

  • cargo build
  • cargo test

v0.1.13

03 Feb 22:30

Choose a tag to compare

Summary

  • align MCP wire format (camelCase), notifications, and resources/read array responses
  • return full GPU process lists in MCP resources
  • enforce guard-mode GPU/group/time policies and user utilization/duration limits
  • normalize global GPU indices across vendors and enrich NVML top process metadata
  • harden coordinator snapshot handling and rogue/whitelist enforcement

Tests

  • cargo build
  • cargo test

v0.1.12

03 Feb 19:05

Choose a tag to compare

Summary

  • fix clippy failure in config tests (avoid field reassign with Default)

Tests

  • not run (relies on CI)

v0.1.11

03 Feb 18:40

Choose a tag to compare

Summary

  • enforce JSON-RPC 2.0 version validation and preserve request IDs in MCP error responses
  • restore config precedence so environment variables override file/defaults
  • include graphics GPU processes in NVML queries with PID de-duplication
  • harden SSH remote argument escaping for Unix targets
  • fix batch kill CLI validation and GPU contention utilization averaging
  • remove unsafe getpwuid username handling on Linux

Tests

  • cargo build
  • cargo test

v0.1.10

31 Dec 02:09

Choose a tag to compare

Release v0.1.10 - Bug fixes

v0.1.9

27 Dec 19:53

Choose a tag to compare

Bump version to 0.1.9 - security patch release

- Fix RUSTSEC-2025-0134 (unmaintained rustls-pemfile)
- Update reqwest to v0.12
- Security audit now passes