fix(transfer): make manifest paths re-exportable

[andrei-hasna]

Summary

• preserve the existing encoded-path selector for normal project and child-path exports
• also match an exact manifest originalPath back to the Claude project directory that produced it
• add a regression for sessions stored under a project root whose metadata names a nested working directory

Root cause

The all-project export resolves originalPath from session metadata. The project-scoped export then tried to find the session directory only by re-encoding that resolved path. When the metadata path was nested below the Claude project directory, the dry run advertised a non-empty project that the real exporter could not select again.

Validation

• regression before fix: 5 pass, 1 fail
• focused transfer lane after fix: 6 pass, 0 fail
• full suite: 379 pass, 0 fail
• bun run typecheck: exit 0
• bun run build: exit 0
• git diff --check origin/main..HEAD: exit 0
• Gitleaks staged and origin/main..HEAD: no leaks found
• Gitleaks positive control: one redacted synthetic AWS-pattern finding
• Shield staged review: no security issues found

Provenance

Todos task: da229805-f8b8-476a-b82a-f414a103dc28

The source-session directories are read only. No sanitizer, copy, import, or source-write behavior changed.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[andrei-hasna]

[REVIEW] GO — #58 @ 6b11edf — lens: correctness+security+gates, reviewer Hostus (1 of 1)

Reviewed the exact one-commit change against origin/main at 16ed634. I read the full diffs for src/lib/transfer.ts and test/transfer.test.ts, then the complete changed files and the surrounding path-resolution helpers, CLI transfer caller, existing path tests, package scripts, and PR rationale.

Commands run:

• bun install — exit 0; setup only, 117 packages installed. This is not reported as a test gate.
• bun run typecheck — exit 0; tsc emitted no pass/fail count.
• bun run test — exit 0; 379 pass, 0 fail, 1798 expect() calls across 44 files.
• gitleaks git --no-banner --redact --log-opts=origin/main..HEAD . — exit 0; 1 commit scanned, no leaks found.

Correctness and security trace: the project selector still preserves encoded exact/child matching. The added path takes the caller-supplied project path through the existing directory inventory, rejects non-directories, compares against the same resolveProjectPath result advertised in the manifest, and then enters the unchanged read/copy export path. It adds no source-session writes, import mutation, authentication boundary, network sink, or permissive failure fallback.

Blocking P0/P1 findings: none.

Non-blocking follow-ups: none.

[andrei-hasna]

[ACCEPTANCE] Station01 real-path verification passed on the exact reviewed candidate 6b11edf48b525cd33fec3ab85b8b9822f0ec9eb1.

Exact sanitized receipt:

{"status":"PASS","installedVersion":"0.12.11","allProjectsForm":"default-without-project","candidateHead":"6b11edf48b525cd33fec3ab85b8b9822f0ec9eb1","advertisedProjects":115,"advertisedNonEmpty":91,"reproducingCandidateIndex":0,"baselineExitCode":1,"fixedExitCode":0,"fixedProjectCount":1,"fixedJsonlCount":1,"fixedTotalFiles":1,"sourceFiles":1,"sourceBytes":124036,"sourceImmutable":true}

OpenLoops run: 019fe59f2530a2acb6265fd4a47b9255.

The installed @hasna/sessions 0.12.11 reproduced the original selection failure on deterministic candidate 0. The frozen PR source exported the same advertised project successfully. Before/after content digests matched, the 0700 staging directory was removed, and no source-session path or body was emitted.

The externally created squash merge 849c724a10e10ec91aabe12a3b978157db69113a has sole parent 16ed6348e0dc991b463f04619004710066367cd9 and tree 6eda749e4fea17563227208f6c23f9350f87f9f2, exactly matching the reviewed head tree.