2.1.24

Changelog - kodustech/kodus-ai

2026-06-20 - 2026-06-26

🐛 Bug Fixes

• #1365 fix: invalidate inherited and scope rules cache after toggling inheri… (@Nalu-devs) ⏳ Not yet released
  #1356
  • Closes #1356: The function to disable inherited rules is not working.
  • 👀 Reviewed by: @kody-ai[bot]

📚 Documentation

• #1364 docs: add June 22 changelog (@edvaldofreitas) ⏳ Not yet released

  Summary - Add the June 22, 2026 changelog release - Add Performance and UX & Interface changelog filters

  • 👀 Reviewed by: @kody-ai[bot]

📦 Other Changes

• #1398 fix(agents): honor BYOK temperature/maxTokens in flow agents (@Wellington01) ⏳ Not yet released
  The flow agents (conversation, business-rules validation, context-evidence) build their LLM call via BaseAgentProvider.createLLMAdapter, which set `.setTemperature(options?.temperature ?? defaultLLM...

  • 👥 Co-authored by: [@claude Opus 4.8 (1M context)](https://github.com/Claude Opus 4.8 (1M context))

• #1393 ci(benchmark): drive tier-0 model benchmark from the QA deploy, not push (@malinosqui) ⏳ Not yet released

  Problem The per-model code-review benchmark (code-review-model-benchmark.yml) has been red on every engine merge for weeks, silently measuring nothing. Root cause is a deploy race, not engine...

  • Closes #1371: feat(code-review): validated review engine — looser finder, soft coverage, cross-file anchoring
  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8 (1M context)](https://github.com/Claude Opus 4.8 (1M context))

• #1392 fix(severity-level-control): standardize colors (@sartorijr92) ⏳ Not yet released
  closes #1391

  • Closes #1391: Standardize the colors in the severity level slicer
  • 👀 Reviewed by: @kody-ai[bot]

• #1389 test(evals): finder-recall eval for the current engine (execution-based replays) (@malinosqui) ⏳ Not yet released

  What A per-stage eval for the current agentic engine so the finder can be iterated in seconds instead of a full benchmark run (hours). Replaces the v2/langchain eval suites.

  • Closes #1385: findFile silently broken in the e2b sandbox: returns fd: command not found instead of file results
  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8 (1M context)](https://github.com/Claude Opus 4.8 (1M context))

• #1387 feat(cockpit): actionable repo & org review reports (@malinosqui) ⏳ Not yet released

  What Replaces the Spotify-style weekly recap with two actionable, data-driven reports: - Repo digest (every 15 days → repo admins): one email per admin, a section per repo they administer.

  • Closes #1382: Repo & org review reports (evolve weekly-recap into actionable digests)
  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8 (1M context)](https://github.com/Claude Opus 4.8 (1M context))

• #1379 fix(e2e): use valid KodyRulesOrigin value in create-rule scenarios (@malinosqui) ⏳ Not yet released

  What PR #1377 (unify kody rules lifecycle) renamed the KodyRulesOrigin enum values — user → manual, and generated → several new values (past_reviews, repo_file_sync, `onboarding_repo_...

  • Closes #1377: feat(rules): unify kody rules lifecycle
  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8 (1M context)](https://github.com/Claude Opus 4.8 (1M context))

• #1377 feat(rules): unify kody rules lifecycle (@jairo-litman) ⏳ Not yet released
  closes #1323

  • Closes #1323: Unify Kody Rules lifecycle, approval, and setup flows
  • 👀 Reviewed by: @kody-ai[bot]

• #1371 feat(code-review): validated review engine — looser finder, soft coverage, cross-file anchoring (@malinosqui) ⏳ Not yet released

  What Bakes the validated finder/loop improvements into the code-review engine as the default behavior. No experiment env flags left — the validated config is now the only config. ### Changes (a...

  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8 (1M context)](https://github.com/Claude Opus 4.8 (1M context))

• #1372 feat(mcp): migrate away composio (@jairo-litman) ⏳ Not yet released
  closes #1182

  • Closes #1182: Plan migration away from Composio MCPs to native/own implementations
  • 👀 Reviewed by: @kody-ai[bot]

• #1328 fix(web): allow MCP OAuth callback after onboarding completes (@a-nasstrom) ⏳ Not yet released

  Summary On self-hosted deployments where onboarding is already complete (finishOnboard=true), the setup layout redirected every /setup/* route to / — including the MCP OAuth callback at `/set...

  • Closes #1327: Self-hosted: MCP OAuth callback redirected before finalize (blocks managed plugins + business-logic)
  • 👀 Reviewed by: @kody-ai[bot]

• #1336 fix(code-review): detect Atlassian Rovo and Jira keys in business logic skip (@a-nasstrom) ⏳ Not yet released

  Summary - Recognize task-management MCP from both mcp_connections and OAuth-active integrations (active on /mcp/integrations), so Atlassian Rovo and custom Jira plugins are detected even when...

  • 👀 Reviewed by: @kody-ai[bot]

• #1331 fix(agents): recognize Atlassian Rovo in business-rules-validation MCP preflight (@a-nasstrom) ⏳ Not yet released

  Summary - Add Atlassian Rovo to required-mcps.examples in business-rules-validation/SKILL.md so MCP preflight accepts Rovo as a task-management provider. - Add a unit test confirming a conn...

• #1369 feat(cockpit): shareable dashboard views via URL filter state (@malinosqui) ⏳ Not yet released

  Why Users couldn't link people to a specific Cockpit view: the dashboard's filter state lived only in cookies + in-memory React state, never in the URL. A shared link always landed on the recipient...

  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8 (1M context)](https://github.com/Claude Opus 4.8 (1M context))

• #1367 fix(github-integration): repair OAuth callback for reset-and-readd flow (@sartorijr92) ⏳ Not yet released
  Closes #1289

  • Closes #1289: GitHub integration cannot be re-added after reset
  • 👀 Reviewed by: @kody-ai[bot]

• #1360 Update README.md (@edvaldofreitas) 📦 Released in 2.1.23