Skip to content

Synchronize the main branch code to the v1.2.3 branch - #619

Merged
syzsunshine219 merged 22 commits into
v1.2.3from
main
Oct 13, 2026
Merged

syzsunshine219 merged 22 commits into
v1.2.3from
main

Conversation

@syzsunshine219

Copy link
Copy Markdown
Collaborator

Synchronize the main branch code to the v1.2.3 branch

Anson-Se and others added 22 commits September 26, 2026 13:17
…mand

A turn routed as end_topic used to be dropped from L1 capture as a whole,
so feedback, work conventions and instructions sent together with the
closing words never reached L1, user memory or the episode reward.

Only a bare end-topic command ("结束会话" / "不聊了") is still kept out of
L1. Every other end_topic turn now goes through the same capture decision
as a normal turn; the closed episode waits for that decision before
reflection and reward are queued.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every memmy-memory run reports memmy_cli_* events to the cloud analytics
endpoint. resolveCommandIdentity took the second positional word as
command_action unconditionally, which was meant for subcommands such as
`session close`. For search, add, get and delete that word is the search
query, the memory text or the memory id, so it was sent verbatim. An
unrecognized first word was also sent verbatim as command_group.

command_action is now reported only when the first word is a group with
subcommands (service, session, turn, raw) and the second word is one of
that group's known subcommands; raw methods match case-insensitively, as
runRaw does. An unrecognized first word is reported as command_group
"unknown". The known commands live in a small table in analytics.ts that
mirrors runCommand and mapTopLevelCommand. has_session_id and has_turn_id
are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The session close route only woke the auto worker when the close result
listed closed episodes, jobs, or scheduled evolution. For an L3 World
Model v2 session, closeSession also freezes the pending L3 evidence
batches and queues l3_world_model_update jobs, but it does not return
them. When the session's episode had already closed earlier (idle
timeout or end_topic), closedEpisodeIds was empty and nothing woke the
worker. The worker does not poll, so the new L3 jobs stayed queued until
an unrelated request happened to wake it.

The close route now always schedules the worker, as the token compaction
boundary already does after freezing batches. A close with no follow-up
work costs one empty drain. A new REST contract test closes an L3 v2
session whose episode had already ended and checks that the worker picks
up the frozen L3 job.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
fix(memory): wake the worker after every session close
v1.2.2 requires the turn capture decision to return a non-empty l1.title
together with l1.summary. The mixed feedback-and-closing test stub only
returned a summary, so the capture decision threw and the closing L1 stayed
in resolving instead of being activated.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
fix(memory): stop sending CLI arguments as analytics command actions
fix(memory): stop dropping end-topic turns that carry feedback from L1
Synchronize the latest code from the main branch to v1.2.2
fix: 兼容 LM Studio 的 JSON 响应格式
…1.2.2

fix(memory): replace stale runtimes by build identity
fix: 限制 Memory 初始化等待,避免会话卡在思考中
chore(release): prepare Memmy v1.2.2
@syzsunshine219
syzsunshine219 merged commit 11c5ef9 into v1.2.3 Oct 13, 2026
9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants