[nateEc]

CodeBuddy compresses normalized workspace directory names above 255 UTF-8 bytes. TeamAI previously used the entire normalized path, so listing/reading missed native sessions and migration could fail with an overlong filename.

Match compressWorkspacePathName from the publisher's CodeBuddy 2.173.0 package: a code-point-safe 180-byte prefix plus - and the unsigned djb2 base36 hash of every normalized byte. Names at or below 255 bytes remain unchanged.

Validation:

• Evaluated fixed vectors directly with the extracted publisher bundle, independently of TeamAI's encoder. Five new long-name vectors failed on main; the exact 255-byte boundary already passed.
• npx vitest run src/__tests__/buddy-windows-cwd.test.ts: 10 passed, covering 255/256 bytes, Unicode/emoji boundaries, distinct suffixes and Windows drive paths.
• npm run build and npx vitest run --config vitest.e2e.config.ts src/__tests__/e2e/buddy-session-directories.test.ts: 2 passed through the built CLI: migrate a Claude transcript into short/long native CodeBuddy directories and read it back into Claude.
• CI exposed a fixture-length assumption on Linux: its short temporary root produced a 244-byte name. Reproduced locally with TMPDIR=/tmp (252 bytes on macOS). The final fixture's segments alone now exceed 255 bytes. After rebuilding, the two CLI cases passed with retries disabled under both the native temp root and TMPDIR=/tmp (2 passed in each run); the 10 encoder unit cases, TypeScript and lint also passed on the final source.
• npx tsc --noEmit, npm run lint, and git diff --check: passed.

State review: both CodeBuddy CLI and WorkBuddy adapter directory lookup and session writes already call the shared encoder; no parallel path implementation or new storage format is introduced. The bilingual session guide and troubleshooting skill data are updated. Windows path seams were tested; native Windows and a live WorkBuddy desktop application were not available locally.

Broader regression: npx vitest run --maxWorkers=6 in a frozen integration worktree containing the five changes finished with 500 files passed, 12,285 tests passed, 36 skipped, 2 failed. Both failures are the pre-existing submodule-checkout.test.ts empty legacy hook-file cases; a clean 83f6d1db worktree independently reproduced 24 passed / the same 2 failed. This is an integration-suite record, not a claim that the baseline is fully green.

Refs #1221. One independently reviewable defect from the issue; the audit/issue remains open for its other items.

[github-actions]

• [P1 blocking] The “long” E2E fixture is too short on Linux with /tmp (src/__tests__/e2e/buddy-session-directories.test.ts:15). For a home such as /tmp/teamai-buddy-e2e-ABC123, the normalized target name is only 246 bytes, so the toBeGreaterThan(255) assertion fails and compression is never exercised. Lengthen the nested segments so the normalized name reliably exceeds 255 bytes.

The description documents sufficient build and real-CLI testing; no missing-testing finding applies. However, the claimed two passing E2E cases conflict with this fixture on a standard Linux temporary directory.

Review was read-only; no PR code was run, built, or installed.

[github-actions]

No new findings in the specified diff.

• Resolved: The earlier [P1 blocking] fixture-length issue is fixed. The nested segments alone exceed 255 normalized UTF-8 bytes, so the long-path case exercises compression even under /tmp.
• Testing: The PR description documents sufficient build and real-CLI verification, including both E2E cases passing after the fixture fix. No missing-testing finding applies.

Review was read-only; no PR code was run, built, or installed.

[SaulMoro]

This was generated by AI during triage.

@nateEc Thanks, approving. This closes the #1221 item on CodeBuddy's long project directory names.

We ported compressWorkspacePathName from CodeBuddy's 2.172.1 bundle (2.173.0 was too new to install here) and compared it with this PR's encoder on 4,000 generated paths, 1,334 of them over 255 bytes, multi-byte characters included. There were no mismatches. With the real CLI, a long-path session round-trips on this head, while the merge-base and main fail with ENAMETOOLONG. tsc, lint and the new tests pass.

Suggestions:

• [P2 non-blocking] The troubleshooting paragraph sits under "Pull left an instruction file unchanged" rather than ### WorkBuddy. The dashboard.md sentences (EN and zh-CN) describe internals a user can't act on. One actionable line, or none, would follow AGENTS.md's "change docs rarely".
• [P3 nit] The fs.ts comment says it hashes "every original byte", but the code hashes the normalized string. The zh-CN text has a stray space after "。". The e2e long case repeats the djb2 loop where a literal would do, since the unit vectors already cover it.

CI review: it reported no new findings and accepted the testing record. We agree on the fix and the record. The two doc points above are the only additions. CI jobs pass at this head.