0.6.14

Highlights

• Fix OpenWebUI preprocessor valve handling and debug behavior
• Add optional trace output to example integrations (off by default)

Changes

• fix: correct OpenWebUI preprocessor valve handling and debug paths
• feat: expose optional trace output in OpenWebUI and LiteLLM examples
• test: expand coverage for host_support helpers and integration edge cases
• docs: add minimal repository meta files
• chore: bump version to 0.6.14 and update example requirements

Notes

No core engine changes
No API changes
LiteLLM proxy trace deferred to future work

v0.6.13

Highlights

• Hardened example integration behavior across LiteLLM and OpenWebUI examples.
• Deterministic update acknowledgments improved:
  • replacement updates now summarize as State updated: Use <new>.
  • policy item acknowledgment casing normalized consistently
  • clear premise now returns Premise cleared.
• Preprocessor example variants now skip fallback LLM calls for directive-shaped malformed inputs, preserving deterministic clarify behavior.
• Added focused regression coverage for replacement acknowledgments, casing consistency, fallback suppression, clear-premise acknowledgments, and near-miss directive clarify behavior.
• Release metadata finalized:
  • package version bumped to 0.6.13
  • OpenWebUI pipe frontmatter versions bumped to 0.7
  • uv.lock updated

v0.6.12 — example integration consistency and host_support API exposure

What's Changed

• Make OpenWebUI example updates deterministic and clarify docs
• Make LiteLLM example updates deterministic
• Expose host_support public API via all and align example imports
• Bump version to 0.6.12

Notes

• Example integrations now render Decision.update deterministically (no downstream LLM call)
• Engine behavior and semantics are unchanged

Full Changelog: v0.6.11...v0.6.12

v0.6.11

What's Changed

Improvements

• Suppress confirmation-only turns from calling the downstream model
• Add deterministic confirmation summaries (e.g., “State updated”, “State unchanged”)

Open WebUI

• Improved BASE_MODEL_ID setup clarity
• Clearer model-not-found error messaging with actionable guidance

Internal

• Added host_support coverage
• Enforced pytest in pre-commit

---

Full Changelog: v0.6.10...v0.6.11

v0.6.10: OpenWebUI 0.9 compatibility patch

Highlights

• Fixed OpenWebUI example compatibility with OpenWebUI 0.9.x async user lookup
• Preserved compatibility with OpenWebUI 0.8.x sync user lookup
• Bumped OpenWebUI example frontmatter versions to 0.4
• Updated OpenWebUI integration README compatibility/install guidance

Notes

• Compatibility patch only; no engine, precompiler, or CLI behavior changes

Full Changelog: v0.6.9...v0.6.10

v0.6.9: precompiler hardening and REPL precompiler option

Highlights

• Hardened experimental precompiler classification with reject-first behavior
• Added portable precompiler conformance fixtures and runner coverage
• Added optional REPL flag: --with-precompiler (heuristic + validation only)
• Clarified integration/docs behavior for near-miss passthrough and engine clarify handling

Also includes CLI polish, OpenWebUI 0.3 example bump, and docs/test updates.

Notes

• No engine semantic changes
• No directive grammar expansion
• REPL default behavior remains unchanged unless --with-precompiler is provided

Full Changelog: v0.6.8...v0.6.9

v0.6.8: fixture-family naming cleanup and regression coverage updates

What's Changed

• fix: standardize remaining clarify prompts by @rlippmann in #100
• docs: add user-facing behavior test check to agents guide by @rlippmann in #101
• fix: align replacement-missing-policy prompt to spec by @rlippmann in #102
• refactor: centralize provider mode helper for host integrations by @rlippmann in #103
• docs: harden structured fixture policy by @rlippmann in #104
• chore: rename fixture families by purpose by @rlippmann in #105
• chore: bump release version to 0.6.8 by @rlippmann in #106

Full Changelog: v0.6.7...v0.6.8

v0.6.7: Checkpoint continuation + integration hardening

Highlights

• Added engine checkpoint APIs for serialized continuation support:
  • export_checkpoint() / import_checkpoint(...)
  • export_checkpoint_json() / import_checkpoint_json(...)
• Added checkpoint adoption in main example integrations (LiteLLM + OpenWebUI), including restore-before-step and persist-on-update/clarify behavior.
• Enforced deterministic pending-clarification continuation behavior in preprocessor integration paths (while pending exists, raw user input is passed directly to engine.step(...)).

Checkpoint and State Safety

• export_json() / import_json() remain authoritative-state-only boundaries.
• Checkpoint import now supports pending as absent or null.
• Checkpoint import rejects semantically invalid pending replacement items (empty / normalize-empty values).
• Added all-or-nothing checkpoint restore behavior coverage.

Integration Reliability and UX

• Added end-to-end checkpoint resume coverage for yes / no continuation paths in preprocessor integration variants.
• Added focused user-facing integration error-path coverage in main examples:
  • missing API key/dependency paths
  • malformed provider response shape handling
  • OpenWebUI model-not-found response/exception normalization
• Improved optional-dependency import resilience in integration test surfaces.

Docs and Release Metadata

• Added/expanded checkpoint contract documentation.
• Updated README with concise checkpoint usage guidance and clear export_json vs checkpoint boundary explanation.
• Updated preprocessor docs with the pending-clarification preprocessing bypass rule.
• Updated integration README notes for checkpoint requirements and storage limitations.
• Bumped project version metadata to 0.6.7.

Other Improvements

• Rejected empty normalized policy keys on state import and added coverage/docs.
• Included CI/contract-gate hardening updates merged in this release window.

v0.6.6: Integration installability and quickstart polish

Summary

0.6.6 is a patch release focused on integration installability, docs clarity, and first-run usability.

Highlights

• OpenWebUI examples and install docs are aligned to 0.6.6.
• Minimal OpenWebUI pipe remains the recommended/default path and is documented with runtime-validated guidance.
• Preprocessor OpenWebUI pipe remains optional/experimental, with aligned frontmatter requirement references.
• LiteLLM and LiteLLM Proxy docs now include clearer quickstart paths, required vs optional config guidance, and concise troubleshooting notes.
• Package version metadata bumped to 0.6.6 (pyproject.toml, uv.lock).

OpenWebUI notes

• Minimal pipe frontmatter requirement:
  • context-compiler>=0.6.6
• Preprocessor pipe frontmatter requirement:
  • context-compiler[experimental]>=0.6.6
• OpenWebUI still requires a real backend model/provider configuration (for example Ollama/OpenAI) for model routing.

Scope

This is a docs/examples/installability patch release.
No core engine behavior redesign is included.

v0.6.5

Summary

Patch release focused on OpenWebUI integration installability/docs polish.

What changed

• Added frontmatter dependency requirement to the minimal OpenWebUI pipe:
  • requirements: context-compiler>=0.6.5
• Updated OpenWebUI integration docs to match 0.6.5:
  • default URL-import flow
  • explicit Docker/manual-install fallback commands
  • model-id selection guidance
• Bumped package version to 0.6.5 (metadata + lock alignment).

Why this release

0.6.4 was close, but the minimal OpenWebUI pipe lacked frontmatter dependency declaration, which made URL-import auto-install less reliable.
0.6.5 closes that gap with a narrow follow-up.

Notes

• Minimal OpenWebUI pipe remains the recommended/default path.
• Preprocessor pipe remains optional/experimental.
• No core engine behavior changes in this patch.