|
| 1 | +## [0.1.6] - 2026-05-14 |
| 2 | + |
| 3 | +### Added |
| 4 | + |
| 5 | +- `thinking_level` profile configuration for local CLI-agent drivers, including Codex `low`, `medium`, `high`, `extra_high`/`xhigh` values and Claude Code `low`, `medium`, `high`, `xhigh`, and `max` values. |
| 6 | +- Protocol-loop coverage for Codex and Claude Code thinking-level propagation. |
| 7 | + |
| 8 | +### Changed |
| 9 | + |
| 10 | +- `codex_exec` now passes configured thinking levels through Codex `model_reasoning_effort`. |
| 11 | +- `claude_code_print` now passes configured thinking levels through Claude Code `--effort`, includes the configured model, uses `--input-format text`, and forwards request JSON schemas with `--json-schema`. |
| 12 | +- `claude_code_print` now reports `json_schema` structured-output support. |
| 13 | + |
| 14 | +### Fixed |
| 15 | + |
| 16 | +- `claude_code_print` now writes the combined prompt to stdin instead of passing it as an argv value. |
| 17 | +- Claude Code local-agent responses now accept `result` as either a JSON string or a JSON object. |
| 18 | + |
| 19 | +## [0.1.5] - 2026-05-09 |
| 20 | + |
| 21 | +### Added |
| 22 | + |
| 23 | +- Support for `bitloops_refresh_cache` request metadata, which sends `X-Bitloops-Refresh-Cache: true` to OpenAI-compatible chat-completions endpoints. |
| 24 | +- Unit coverage for explicit cache-refresh headers and one-time Bitloops platform cache-refresh retry decisions. |
| 25 | + |
| 26 | +### Fixed |
| 27 | + |
| 28 | +- Bitloops platform JSON-object parse failures now retry once with cache refresh before returning an invalid provider response. |
| 29 | + |
| 30 | +## [0.1.4] - 2026-05-04 |
| 31 | + |
| 32 | +### Added |
| 33 | + |
| 34 | +- Structured-generation profile support alongside text-generation profiles. |
| 35 | +- Local CLI-agent providers for `codex_exec` and `claude_code_print`, including provider metadata, runtime command/argument handling, workspace-path support, JSON-schema handling, and normalized JSON responses. |
| 36 | +- `structured_output` provider capabilities in the protocol metadata. |
| 37 | +- Default runtime startup and request timeouts for CLI-backed profiles. |
| 38 | +- An ad hoc `bitloops-inference-perf` runner for live provider latency, throughput, and token-usage reporting. |
| 39 | +- Documentation for structured local CLI-agent profiles and the manual performance runner. |
| 40 | + |
| 41 | +### Changed |
| 42 | + |
| 43 | +- Config loading now accepts both underscore and hyphen spellings for supported profile tasks. |
| 44 | +- Config validation now checks task/driver compatibility and ignores daemon profiles unrelated to text or structured generation. |
| 45 | + |
1 | 46 | ## [0.1.3] - 2026-04-17 |
2 | 47 |
|
3 | 48 | ### Added |
|
0 commit comments