* test: make ClawScan process-tree timeout test deterministic
The timeout test raced its 500ms deadline against the fixture writing
descendant.pid, and treated zombies as live processes via kill(pid, 0).
Under parallel coverage runs it flaked. Now waits for the pid barrier,
drives the timeout with fake timers, and treats zombie state as exited.
* ci: retry transient convex preview provisioning failures
Fresh Convex preview deployments intermittently 404 on get_config_hashes
while provisioning, failing the whole Vercel preview build after the
CLI's internal retries. Retry the preview deploy step up to 3 attempts
with 20s/40s backoff; other steps keep fail-fast behavior.
* fix: restore promotion bar icon geometry token
77459acc dropped border-radius: var(--oc-radius-inset) from
.promotion-bar-icon while folding the removed fallback rule into it,
breaking the ui-design-contract test on main.
* ci: retry preview pipeline under fresh preview names
Retrying --preview-create under the same name leaves two deployments and
convex run --preview-name can resolve to the dead one (seen live: seed
failed with missing functions after a successful retry). Each retry now
reruns deploy plus seed under <branch>-retry-N so resolution is unique.
* fix: keep CLI device codes out of the OAuth code handler
The global AuthCodeHandler consumed any ?code= query param as a GitHub
OAuth completion code. CLI device login links (/cli/device?code=XXXX-XXXX)
hit that path: the device code was stripped before the page could read it,
the failed code exchange erased the active session, and the retry logic
bounced users through a surprise GitHub redirect.
Device verification links now use user_code, the OAuth handler ignores
device-shaped codes as defense in depth, and the device page accepts the
legacy param only when it matches the device code format.
* chore: refresh stale convex generated api for skillTags
Accept legacy skill verify --json usage as a hidden compatibility no-op, add a regression e2e for the flattened verifier response, and prepare clawhub@0.19.2.
* fix: restrict membership management to org publishers
* fix(api): ignore stale personal publisher memberships
* fix(api): reject stale personal publisher publish targets
* fix(api): reject stale personal publisher memberships
* fix(api): use personal publisher links for package access
* fix(api): guard personal publisher owner scopes
* fix: enforce publisher ownership for skill reads
* fix: narrow personal publisher dashboard owner
* fix: ignore stale personal package memberships
* fix: allow own legacy personal skill destination
* fix: preserve legacy personal package dashboards
* fix: preserve legacy personal skill dashboards
* fix: avoid redundant personal publisher boolean coercion
* fix: preserve legacy personal publisher access
* fix: include legacy direct packages in personal dashboard
* fix: close stale personal publisher ownership gaps
---------
Co-authored-by: Patrick Erichsen <patrick.a.erichsen@gmail.com>
* fix: hide package resources when owners are banned
* fix(api): attribute unban package restores to moderator
* fix(web): clarify package effects in ban confirmations
* fix(api): continue package ban batches during in-flight bans
* fix: keep package unban restore scoped to ban batch
* fix: retimestamp package releases during repeated bans
* fix: start package ban batches after user ban commit
* fix: preserve manual package moderation after unban
* fix: cover personal publisher package sanctions
* fix: restore personal publisher packages in autoban remediation
* fix: bound package publish token revocation batches
* fix: clear package ban reason during remediation restore
* fix: block direct package ban restores
* fix: scan linked personal publisher packages during sanctions
* fix: tighten package sanction restore batches
* fix: block personal publisher publishes after owner ban
* fix: allow initial package ban cleanup before commit
---------
Co-authored-by: Patrick Erichsen <patrick.a.erichsen@gmail.com>
* fix: block direct skill transfers under moderation
* fix(api): block accepted transfers for moderated skills
* fix(api): block malware-flagged skill transfers
* fix: cover moderated skill transfer bypasses
* test: support ownership heal transfer sync
* fix: close moderated transfer backfill gaps
* fix: block soft-deleted transfer guard state
---------
Co-authored-by: Patrick Erichsen <patrick.a.erichsen@gmail.com>
* fix: prevent skill transfer acceptance after requester is banned
The acceptTransferInternal mutation did not verify whether the transfer
requester (fromUser) was banned or deactivated when the skill still
belonged to that user. This created a race condition where a pending
transfer could be accepted after the requester was banned, allowing
the skill to escape the ban batch and remain alive under a new owner.
This change moves the requester validity check before the ownership
branch, so it is evaluated unconditionally for all transfers.
Fixes a security vulnerability where banned users' skills could
survive moderation actions via pending transfers.
* fix: harden skill transfer acceptance
Co-authored-by: vyctorbrzezowski <krzyszchweski@gmail.com>
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
Vincent Koc