| feat(streaming): mid-stream retry with checkpoint continuation
streamModel now retries up to 2 times on transient network errors
(ECONNRESET, socket hang up, etc). When substantial partial content
has been received, it builds a continuation request to resume from
the checkpoint rather than starting over.
Co-authored-by: Cursor <cursoragent@cursor.com>
| 20 天前 |
| feat(model): extract tool calls from raw text for non-conformant models
Add parseTextToolCalls module that recognizes Qwen XML, DeepSeek DSML,
Hermes JSON-in-XML, Mistral [TOOL_CALLS], and Llama python_tag formats.
Wire it into assembleModelMessage as a fallback when no structured tool
calls are present. Add PILOTDECK_DUMP_REQUEST env for request debugging.
Co-authored-by: Cursor <cursoragent@cursor.com>
| 15 天前 |
| feat: add tool and agent runtime foundations
Build the PolitDeck tool/permission runtime and start the agent loop on canonical model/tool contracts so future refactors can prove parity against legacy behavior.
Co-authored-by: Cursor <cursoragent@cursor.com>
| 26 天前 |
| feat(model): extract tool calls from raw text for non-conformant models
Add parseTextToolCalls module that recognizes Qwen XML, DeepSeek DSML,
Hermes JSON-in-XML, Mistral [TOOL_CALLS], and Llama python_tag formats.
Wire it into assembleModelMessage as a fallback when no structured tool
calls are present. Add PILOTDECK_DUMP_REQUEST env for request debugging.
Co-authored-by: Cursor <cursoragent@cursor.com>
| 15 天前 |
| feat(model): extract tool calls from raw text for non-conformant models
Add parseTextToolCalls module that recognizes Qwen XML, DeepSeek DSML,
Hermes JSON-in-XML, Mistral [TOOL_CALLS], and Llama python_tag formats.
Wire it into assembleModelMessage as a fallback when no structured tool
calls are present. Add PILOTDECK_DUMP_REQUEST env for request debugging.
Co-authored-by: Cursor <cursoragent@cursor.com>
| 15 天前 |