✨ Feat: Context management refactoring. Add benchmark for agent context module.#3039

Open

JasonW404 wants to merge 34 commits into

ModelEngine-Group:developfrom

liudfgoo:feat/benchmark-on-refactor

Commits on May 20, 2026

♻️ refactor: consolidate context building into ContextManager
Jason
committed

Commits on May 27, 2026

fix(db): add missing semicolon to is_new index in init.sql

wuyuanfr
and
claude
committed
♻️ refactor: route rendered system_prompt through build_context_components

wuyuanfr
and
claude
committed
🧪 test: add prompt-equivalence golden test

wuyuanfr
and
claude
committed
🔧 chore: log when a ContextStrategy drops a component

wuyuanfr
and
claude
committed
✨ feat(sdk): pass-through extra_body to chat.completions.create

wuyuanfr
and
claude
committed
✨ feat(sdk): pre-truncate long observations under ContextManager

wuyuanfr
and
claude
committed
✨ feat(sdk): add benchmark instrumentation APIs on ContextManager

wuyuanfr
and
claude
committed
📦 chore: import sdk/benchmark and sdk/ctx_debugger from feature branch

wuyuanfr
and
claude
committed
🧪 test: add minimal end-to-end smoke for benchmark-on-refactor

wuyuanfr
and
claude
committed
🐛 fix(benchmark): r.get is a method not a subscriptable

wuyuanfr
and
claude
committed
🐛 fix(test): allow prompt_type kwarg in compress side_effect mocks

wuyuanfr
and
claude
committed
🧪 fix(test): restore real sys.modules after stub-injecting tests

wuyuanfr
and
claude
committed
♻️ refactor(sdk): implement piecewise component assembly for all 4 variants
wuyuanfr
committed
✨ feat(sdk): add compress_history_offline standalone function

wuyuanfr
and
claude
committed
✨ feat(sdk): pass-through max_tokens to chat.completions.create

wuyuanfr
and
claude
committed
📝 docs: translate Chinese markdown files to English in benchmark and ctx_debugger
wuyuanfr
committed
🔄 sync(benchmark): pick eventqa_eval + agent_runner from feature

wuyuanfr
and
claude
committed
sync(benchmark): pick run_longmemeval + run_with_debugger+ summary_schemas from feature
wuyuanfr
committed
🔄 sync(benchmark): pick manual_cases harness from feature

wuyuanfr
and
claude
committed
🔧 chore(benchmark): align build_agent_run_info max_tokens default to None

wuyuanfr
and
claude
committed
update system and summary prompt

liudfgoo
authored and
wuyuanfr
committed
readme and eval res

liudfgoo
authored and
wuyuanfr
committed
add fallback and build_compressed_snapshot

liudfgoo
authored and
wuyuanfr
committed
add acon output res

liudfgoo
authored and
wuyuanfr
committed
add default summary prompt argument

zooowooxo
authored and
wuyuanfr
committed
add cases

zooowooxo
authored and
wuyuanfr
committed
add reports results

zooowooxo
authored and
wuyuanfr
committed
add static summary inspections

zooowooxo
authored and
wuyuanfr
committed
Merge branch 'develop' into feat/benchmark-on-refactor
JasonW404
committed
fix(sdk): restore indentation on extra_body after merge
wuyuanfr
committed
fix(sdk,test): resolve merge fallout in AgentConfig and align tests
wuyuanfr
committed

Commits on May 28, 2026