Популярные репозитории
- Python
- Python
37 активностей за последний год
- b5045d1
critic: JSONL telemetry for post-LLM checks + report script (V13a)
- 1294b11
critic+runner+batch: hybrid anti-loop — hash + LLM semantic confirm (V12)
- b1ee0d9
writer: cap→60k + LLM-summary fallback, never truncate critic_feedback (V15)
- c6fe5a4
meta: 2 LLM merge attempts (full → shrunk prompt) before Python-union (V14)
- d4fce3a
writer+runner+batch: propagate parse-fail reason to next critic_feedback (V11)
4 месяца назад- a0f0d75
runner+batch: preserve case in canonical_tc hash (whitespace-normalize only)
- a775e24
writer: rotate parse-fail debug dir (keep last 100)
- ae5697c
meta: Python-union fallback when MERGE LLM fails (preserve unique NEW)
- cfc45c4
map_context: expose up to 10 good_patterns (was 3)
- 64298fe
batch: empty steps → continue (symmetry with runner)
4 месяца назад- c6c7a9f
writer: enforce good_patterns compliance in SYSTEM_PROMPT
- 35622a0
writer: cap prompt size with truncation fallback
- adcb110
writer: retry escalation — forbid identical response on iter>=2
- cfb4af2
writer: anti-loop detection via canonical TC hash in runner/batch_runner
- f2c841b
writer: graceful parse — save raw to debug on extract_json failure
4 месяца назад- f81d9cd
meta: save orphan result when project_map missing
- 4a22ef6
meta: circular-learning guard — good_patterns source strictly team TCs
- 312c9b8
meta: move score_distribution out of qa_patterns to top-level
- 3b03f2a
meta: qa_standards guard + history symmetric to qa_patterns
- 192c1a7
meta: LLM-powered merge of qa_patterns (replaces overwrite)
4 месяца назад- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад
- 4 месяца назад