2.1.35

Changelog - kodustech/kodus-ai

2026-07-31 - 2026-08-05

📚 Documentation

• #1654 docs: fix code/docs drift across the external docs (@malinosqui) ⏳ Not yet released

  Summary Audited the published docs against this branch and fixed what no longer matches the code. Everything here is either a factual correction, a shipped-but-undocumented feature, or documentatio...

  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 5 (1M context)](https://github.com/Claude Opus 5 (1M context))

📦 Other Changes

• #1658 fix(github): fail fast + requeue on rate limit instead of sleeping silently (+ split e2e account by license) (@sartorijr92) ⏳ Not yet released

  What this replaces Reverts #1657 (first commit) and ships the correct fix. #1657 chased a wrong "review is LLM-slow" hypothesis (lowered the agent timeout, reworked the e2e retry gate). The rea...

  • Closes #1657: fix(code-review): keep slow reviews from reading as "no review" + surface LLM throttling
  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8](https://github.com/Claude Opus 4.8)

• #1657 fix(code-review): keep slow reviews from reading as "no review" + surface LLM throttling (@sartorijr92) ⏳ Not yet released

  Problem A @kody review <directive> (focus) review runs the finder to completion, but its hard per-agent ceiling (AGENT_TIMEOUT_MS = 30 min) sat above the E2E command-review poll window (`15...

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

• #1655 feat(cli): hunk 0.18, findings sidebar, readable notes, worktree fix (@malinosqui) ⏳ Not yet released

  Why Three things, all found or confirmed by running the CLI for real rather than from tests alone. The CLI did not work in a git worktree. kodus hook install crashed with ENOTDIR and the ot...

  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 5 (1M context)](https://github.com/Claude Opus 5 (1M context))

• #1653 fix(agent-harness): preserve tool-result output shape when compressing (@sartorijr92) ⏳ Not yet released

  Problem When a review runs on a large PR, the main model (an OpenAI Responses-API model) fails every generation with: ```

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

• #1648 fix(kody-rules): trace the sharded judge in Langfuse (@sartorijr92) ⏳ Not yet released

  Problem The kody-rules review agent (deterministic sharded path, #1449) produces zero traces in Langfuse, while every other review agent traces normally. Root cause: the agent's `execut...

  • Closes #1449: kody-rules under-covers: it inherits the bug-finder's depth-first prompt (a breadth task on a depth brain)
  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8](https://github.com/Claude Opus 4.8)

• #1644 fix(kody-rules): load @file: references into the sharded review (@sartorijr92) ⏳ Not yet released
  Closes #1643 ## Problem When a Kody Rule cites a repository file in its body via an @file: marker,

  • Closes #1643: Kody Rules: @file: references in rule body are not loaded into the review (sharded path ignores contextReferenceId)
  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8](https://github.com/Claude Opus 4.8)

• #1639 fix(billing): trial-specific status + BYOK divergence observability (@sartorijr92) ⏳ Not yet released
  Closes #1556

  • Closes #1556: Trial org with BYOK registered in billing gets blocked by trial-credit gate — BYOK state diverges between billing and organization_parameters
  • 👀 Reviewed by: @kody-ai[bot]
  • 👥 Co-authored by: [@claude Opus 4.8](https://github.com/Claude Opus 4.8)

👥 Contributors

• @sartorijr92
• @kody-ai[bot]
• [@claude Opus 4.8](https://github.com/Claude Opus 4.8)
• @malinosqui
• [@claude Opus 5 (1M context)](https://github.com/Claude Opus 5 (1M context))