<@U0BME18CF7S> look at all tests that use ~ or use...
# proj-agent
c
@fresh-machine-37269 look at all tests that use ~ or user home; if they persist things like profiles or settings in user’s home dir, fix them to use the persistence dir env var for /tmp or something (you can probably find good examples, fix the bad examples, if any) β€” in repo openhands/agent-sdk open pr, non-draft
πŸ‘€ 1
f
πŸ€– On it! View progress here: oss-agent-canvas.ngrok.dev/conversations/…
⏳ Still working β€” this is taking longer than this run allows. Follow along at the link above; mention me again for an update.
m
@fresh-machine-37269 can you give an update? Moreover, why are you so slow?
πŸ‘€ 1
f
πŸ€– On it! View progress here: oss-agent-canvas.ngrok.dev/conversations/…
βœ… Done! Here's the update on the `~`/home-dir persistence cleanup for `openhands/software-agent-sdk`: Two open, non-draft PRs already cover this work: 1. #4476 β€”
fix: respect OH_PERSISTENCE_DIR for all ~/.openhands paths
(the actual fix). It centralizes every
Path.home() / ".openhands"
call site behind a single
get_user_persistence_dir()
helper, so profiles/settings/credentials/caches respect
OH_PERSISTENCE_DIR
, plus a 430-line cross-cutting test that exercises every call site. This is the substantive change β€” but it's currently blocked with
CHANGES_REQUESTED
: a maintainer (Vasco) objected to adding the new
<MEMORY>
system-prompt section, and that feedback hasn't been addressed yet. It's mergeable but stuck on review. 2. #4521 β€”
test(agent-server): isolate MCP OAuth settings persistence via OH_PERSISTENCE_DIR
(a small test-isolation fix). Points
OH_PERSISTENCE_DIR
at
tmp_path
in the three
test_mcp_oauth_store.py
tests so encrypted OAuth token state stops leaking into the host home dir. Awaiting review (
REVIEW_REQUIRED
). On the "why so slow" part: the work itself is done and sitting in PR form β€” the bottleneck isn't the code, it's that #4476 has been parked on a maintainer's changes-request since Aug 13 and nobody's resolved the system-prompt disagreement. So the actionable next step is to either (a) drop/adjust the new
<MEMORY>
prompt section and re-request review, or (b) reply to Vasco's thread to reach a decision. Want me to take a crack at resolving the review feedback and push an update to #4476?
c
Ah, is 4476 okay with you @modern-fireman-99755 ?
m
What do you mean? for me is good. There are just a couple of fix to adress but it seems fine to me
c
Ah, I see. The rest seem nits, thank you cc @curved-angle-80449