Trajectory manager migration v4 - #2

Open

jingshenghang wants to merge 28 commits into

mainfrom

trajectory-manager-migration-v4

Commits on Jun 8, 2026

refactor(agent): replace segment-based trajectory with turn-node TrajectoryTree
jingshenghang
committed
8646a8d
refactor(agent): drop dead code left by trajectory-manager refactor
jingshenghang
committed
f10552f
refactor(agent): port anthropic + trajectory_manager from trajectory-manager-migration-v2
jingshenghang
committed
ba557ae
refactor(agent): rewrite openai.py for Codex CLI + TrajectoryManager
jingshenghang
committed
b8a5cd4
refactor(agent): centralize snapshot-threshold default + filter access log
jingshenghang
committed
44e7999
feat(agent): add fork-merge rescue for short assistant rewrites
jingshenghang
committed
5e59c25
fix(agent): replace sib.messages on fork-merge rescue
jingshenghang
committed
1d459bd
refactor(agent): drop billing-header scrub now cc emits no header
jingshenghang
committed
8d6fdc9
refactor(agent): migrate TrajectoryManager and adapters (v4)
jingshenghang
committed
2b4efc4
refactor(agent): drop drift fork/merge params, strict exact-prefix linearization
jingshenghang
committed
6f95f18
feat(agent): assistant-rewrite merge to de-dilute reward
jingshenghang
committed
4fcbb24
feat(agent): TrajectoryManager re-accepts fork_merge_max_response_tokens
jingshenghang
committed
0f1c5aa
docs(test): spec for TrajectoryManager e2e test script
jingshenghang
committed
4732702
test(agent): end-to-end TrajectoryManager test matrix (append_turn/get_trajectory)
jingshenghang
committed
4ea5d5a
test(agent): dump raw append_turn inputs in e2e readable output
jingshenghang
committed
cbee0de
test(agent): 1.7 now shows token drift's effect on the linearized sample
jingshenghang
committed
a11c9b4
test(agent): every case prints [samples] + mask info
jingshenghang
committed
f1b1792
test(agent): make reward-split explicit in dump + conservation assert
jingshenghang
committed
0be5966
test(agent): set every case's input reward to 1.0
jingshenghang
committed
fe7692a
test(agent): render whitespace in token labels as visible ␣
jingshenghang
committed
2ee5ee8
test(agent): add Group 4 (boundary/defensive/feature) -> 98% coverage
jingshenghang
committed
8eed4dd
test(agent): assert full output via golden token+loss strings
jingshenghang
committed
cfad29d
refactor(agent): drift-tolerant trajectory linearization
jingshenghang
committed
624b927
chore(agent): untrack e2e test design doc and trajectory_manager tests
jingshenghang
committed
054f89a
chore(agent): drop comments/docstrings in generate.py and TurnRecord
jingshenghang
committed
bc3d304
docs(agent): tighten trajectory_manager comments to why-not-what
jingshenghang
committed
ece9007

Commits on Jun 9, 2026

refactor(agent): slim adapters and trajectory_manager, add e2e test
jingshenghang
committed
6565fe1
refactor(agent): assert base_sample in get_trajectory instead of defaulting
jingshenghang
committed
ad122f4

Navigation Menu

Trajectory manager migration v4 - #2

Trajectory manager migration v4#2
jingshenghang wants to merge 28 commits into
mainfrom
trajectory-manager-migration-v4

Commits on Jun 8, 2026

refactor(agent): replace segment-based trajectory with turn-node TrajectoryTree

refactor(agent): drop dead code left by trajectory-manager refactor

refactor(agent): port anthropic + trajectory_manager from trajectory-manager-migration-v2

refactor(agent): rewrite openai.py for Codex CLI + TrajectoryManager

refactor(agent): centralize snapshot-threshold default + filter access log

feat(agent): add fork-merge rescue for short assistant rewrites

fix(agent): replace sib.messages on fork-merge rescue

refactor(agent): drop billing-header scrub now cc emits no header

refactor(agent): migrate TrajectoryManager and adapters (v4)

refactor(agent): drop drift fork/merge params, strict exact-prefix linearization

feat(agent): assistant-rewrite merge to de-dilute reward

feat(agent): TrajectoryManager re-accepts fork_merge_max_response_tokens

docs(test): spec for TrajectoryManager e2e test script

test(agent): end-to-end TrajectoryManager test matrix (append_turn/get_trajectory)

test(agent): dump raw append_turn inputs in e2e readable output

test(agent): 1.7 now shows token drift's effect on the linearized sample

test(agent): every case prints [samples] + mask info

test(agent): make reward-split explicit in dump + conservation assert

test(agent): set every case's input reward to 1.0

test(agent): render whitespace in token labels as visible ␣

test(agent): add Group 4 (boundary/defensive/feature) -> 98% coverage

test(agent): assert full output via golden token+loss strings

refactor(agent): drift-tolerant trajectory linearization

chore(agent): untrack e2e test design doc and trajectory_manager tests

chore(agent): drop comments/docstrings in generate.py and TurnRecord

docs(agent): tighten trajectory_manager comments to why-not-what

Commits on Jun 9, 2026

refactor(agent): slim adapters and trajectory_manager, add e2e test

refactor(agent): assert base_sample in get_trajectory instead of defaulting