Fix stale token winning deduplication race in LevelDB extraction

[devxoul]

Summary

• The dedup logic let stale tokens from old .ldb files overwrite fresh tokens when the stale entry had a team name but the fresh one didn't. Combined with unsorted .ldb file ordering, this caused intermittent auth failures after LevelDB compaction.

Changes

src/platforms/slack/token-extractor.ts

• Change deduplicateTokens() from last-wins to first-wins for token values. A later entry can only upgrade teamName from unknown — the token value from the first (freshest) source is always preserved.
• Sort .ldb files by modification time (newest first) so fresh tokens are encountered before stale ones during scanning.
• Add statSync import for file mtime comparison.

src/platforms/slack/token-extractor.test.ts

• Test that first token wins dedup and team name is upgraded from a later entry.
• Test that .log tokens take priority over .ldb tokens for the same team.

Verified

• bun test — 14 pass, 0 fail.
• bun typecheck — clean.
• bun lint — clean.

---

Summary by cubic

Fixes a dedup race where stale Slack tokens from old .ldb files could overwrite fresh tokens, causing intermittent auth failures after LevelDB compaction. Fresh tokens now consistently win, with file ordering favoring newest data.

• Bug Fixes
  • Changed token dedup to first-wins for token values; later entries can only upgrade an unknown team name.
  • Sorted .ldb files by modification time (newest first) and continued to prioritize .log files to scan fresher tokens first.

^{Written for commit 47eae3e. Summary will update on new commits.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
agent-messenger	Ignored		Mar 5, 2026 0:33am

[cubic-dev-ai]

No issues found across 2 files