Repository navigation
📖 Sessions: single-owner Codex/OpenCode login docs, invariant test, live-install checklist - #18881
Conversation
…Refs #18740) A refresh-less session copy would log the CLIs out, so the adopt-back stays the safety net. Pins that adoption never writes the session copy, documents the reasoning and the manual live-install checklist. Signed-off-by: Andrew Anderson <andy@clubanderson.com>
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
|
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor. @clubanderson this PR is held at Failing checks:
Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path. Evidence bundle `#18881f951f9f` sha256:2519bca03db28f1f4b60d3a5c9045aff9a6578558e65f11dca232eb28a84df69 — download from the dashboard or `hivectl review evidence #18881`.merged by Hive trusted-author auto-merge: author |
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
There was a problem hiding this comment.
Read the PR at head 5081be3 against the tree: the stale-file retention the new docs describe is real (usersession_login_rotating.go:288 replaceSharedLogin -> preserveClaudeCredential, which keeps the replaced shared file as <name>.stale-<UTC time>, and the audit entry carries stale_shared_file at :330). The new test pins what the comment now says: adoption writes only the shared file, and a later agent refresh is not displaced. Looks correct to me; the code change is comment-only.
One docs nit (not blocking): in src/docs/security-model.md the manual checklist's last sentence ("Attach a redacted real file ... as the test fixture.") runs straight into the pre-existing "For all of these vendors a resume no longer overwrites ..." sentence on the same line, so the unrelated resume point now reads as part of the checklist. A line break before "For all of these vendors" would fix it. The double parenthetical "(from their source, not yet confirmed on a live install) (Codex: ...)" also reads awkwardly.
The Codex/OpenCode refresh-token-required claim is, as the PR says, from source rather than a live install, and I did not verify it either; Refs #18740 is right to leave the issue open for the manual check.
— hive: agent=reviewer backend=claude model=claude-sonnet-5-5 claude=2.1.326
Refs #18740 (does not close it: the live-install check needs a person).
Option 2 decision. Not shipped as a code change. Codex (
tokens.refresh_token) and OpenCode (refreshof an oauth entry) keep the refresh token as a required field of their login file (from their sources; unconfirmed on a live install), so a refresh-less or placeholder session copy would sign the person out of the session. Per the issue, #18793's adopt-back stays the safety net. The shared login remains the single owner: adoption only writes the shared file, never the session copy.Changes
auth.json.stale-<time>), audit entry, and a manual live-install checklist.Manual checklist (needs a person): see the security-model.md section; attach a redacted real auth file as a fixture.
Highest risk: the docs claim about CLI schemas is from source knowledge, not a live check.
🤖 Generated with Claude Code