Skip to content

Releases

kkt v0.8.1

Choose a tag to compare

@github-actions github-actions released this 19 Aug 18:37

Binary assets for kkt v0.8.1.

Added

  • Add CP-SAT-inspired constraint-search guidance to $kkt-deep.
  • Add harness-capability-aware read-only delegation with main-agent fallback.

Known limitations

  • KKT does not run a CP-SAT solver.
  • Delegation requires a compatible native capability exposed by the active harness.

kkt v0.8.0

Choose a tag to compare

@github-actions github-actions released this 11 Aug 17:15

Added

  • Add kkt-deep, an end-to-end workflow for architecture decisions, high-impact tradeoffs, and complex implementation work.
  • Add SWE-bench Verified benchmark results comparing vanilla, plan mode, kkt, and kkt-deep.

Changed

  • Consolidate the installed workflows into kkt and kkt-deep; upgrades remove the legacy kkt-model, kkt-run, and kkt-loop skill directories.
  • Replace optional agent hook enforcement with compact execution.json bounds for scope, required validation, approval drift, and stale evidence.
  • Remove the installer’s --hooks option and clean up known legacy KKT hook adapters and configuration during install, upgrade, and uninstall.
  • Make both active workflows approval-gated and validation-driven, with correctness-first minimal changes and bounded read-only modeling delegation in kkt-deep.
  • Keep legacy run and loop CLI commands for compatibility, but no longer package or document them as active skills.

kkt v0.7.0

Choose a tag to compare

@github-actions github-actions released this 29 Jul 03:27

Added

  • Added constraint-driven clarification across KKT workflows. Agents now ground questions in repository facts and a provisional constraint model before implementation planning.

Changed

  • Fresh modeling asks only for owner decisions that affect feasibility or plan selection, while completed imported models skip repeated intake unless material drift is detected.
  • Simplified installation and upgrade guidance: automatic agent detection remains the default, and --target is an explicit override.

Fixed

  • Hardened Pi and OpenCode hook adapters to fail open safely when hook execution is unavailable or returns invalid output.
  • Corrected OpenCode post-tool payload handling and added a bounded hook-execution timeout.

kkt v0.6.0

Choose a tag to compare

@github-actions github-actions released this 21 Jul 20:30

Added

  • Added compact and deep constrained-optimization contracts across KKT workflows.
  • Added direct bounded execution with kkt start run.
  • Added simplified guardrail configuration and richer JSON status output.

Changed

  • Improved workflow validation while preserving compatibility with existing workspaces.
  • Simplified guidance for choosing compact or deep modeling.

kkt v0.5.2

Choose a tag to compare

@github-actions github-actions released this 12 Jul 21:49

Added

  • Added opt-in beta guardrail hook adapters for Codex, Claude Code, Pi, and OpenCode, with observe/enforce modes and automatic disarming.
  • Added kkt loop from-model to start a durable iterative workflow from a completed model.

Changed

  • Standardized KKT planning around one Optimized Plan Contract across compact, model, run, and loop workflows.
  • Workflow transitions now validate their own prerequisites, reducing required diagnostic ceremony.
  • Deprecated kkt start run; use kkt run from-model instead.

Fixed

  • Fixed local installation when Codex has no legacy KKT hook block in config.toml.

Known limitations

  • Hook adapters remain opt-in beta; Pi and OpenCode still require real-runtime validation.

kkt v0.5.1

Choose a tag to compare

@dannylee1020 dannylee1020 released this 08 Jul 03:59

Added

  • Added an npm-distributed installer so users can install or upgrade KKT with npx @dannylee1020/kkt install --target ....
  • Added package metadata for Pi skill/package installation from npm.

Changed

  • Updated install documentation to recommend the npm/npx install path while keeping the curl installer available.
  • KKT guardrails now ignore unrelated dirty files outside modeled allowed paths, keeping validation focused on explicit blocked paths and in-scope changes.

Fixed

  • Fixed validation proof staleness caused by unrelated out-of-scope working tree changes.

kkt v0.5.0

Choose a tag to compare

@github-actions github-actions released this 06 Jul 17:35

Added

  • kkt validate --run records deterministic proof for required validation commands.
  • Installer support for ast-grep so KKT skills can rely on syntax-aware discovery.
  • Direct rg and ast-grep discovery guidance across KKT skills.

Changed

  • kkt status --json now reports validation state, stale-complete state, and next action.
  • Model-to-run handoff now preserves imported model layers and source workspace metadata.
  • Guardrails now distinguish preexisting dirty files from post-approval drift.
  • Skill references were tightened into a leaner shared workflow contract.

kkt v0.4.0

Choose a tag to compare

@github-actions github-actions released this 29 Jun 23:20

Added

  • Added $kkt-run for executing a completed KKT model with deterministic guardrails instead of the long-running loop workflow.
  • Added kkt guardrails, kkt judge, and kkt run from-model CLI flows for guardrail state, checkpoint validation, and model-to-run handoff.
  • Added path-drift enforcement so run and loop workflows can block changes outside modeled allowed paths or inside blocked paths.
  • Added explicit KKT planning contracts that capture objective functions, files to modify, constraint functions, decision variables, and validation proof.

Changed

  • KKT model and loop skills now require model-derived guardrail variables before implementation: constraints, allowed paths, blocked paths, required commands, and validation evidence.
  • Run and loop workflows now use deterministic checkpoints for model readiness, pre-mutation approval/path checks, continuation checks, and final validation.
  • Installer documentation now includes KKT_BUILD_FROM_SOURCE for building the CLI from source instead of downloading a release binary.

Known limitations

  • Guardrail judging is deterministic in this release. It checks workflow state, approval, validation, replay state, stop conditions, and path bounds, but it does not semantically inspect code behavior.

kkt v0.3.3

Choose a tag to compare

@github-actions github-actions released this 29 Jun 03:05

Changed

  • KKT now anchors durable .kkt/ state at the nearest Git/worktree project root, so kkt start, kkt status, and continuation commands resolve the same workspace when run from nested directories.
  • Outside a Git/worktree checkout, KKT keeps the existing fallback behavior and stores .kkt/ under the current directory.
  • README and bundled skill references now document the project-root state location.

kkt v0.3.2

Choose a tag to compare

@github-actions github-actions released this 26 Jun 23:13

Added

  • Added structured kkt next --json output for machine-readable loop continuation.
  • Added kkt resume continuation packets and kkt replay --check consistency checks for durable loop workspaces.
  • Added criterion-mapped validation evidence with optional recorded validation commands.

Changed

  • Strengthened loop validation so completion requires approval, closed tasks, satisfied criteria, and mapped evidence.
  • Clarified loop state docs around kkt.yaml as current state and events.jsonl as append-only audit and resume history.

Fixed

  • Fixed fresh loop kkt next guidance so new loop workspaces continue from discovery instead of jumping to validation.
  • Fixed unapproved loop kkt next --json guidance so pending work requests approval before task execution.