docs(sample-data): mark coding-projects auto-skip + settings.json skip as aspirational

[shiba4life]

Summary

• Reconciles sample_data/README.md with actual smart-folder behavior. The README claimed manifest-based auto-skip (package.json / Cargo.toml / pyproject.toml) and config-file skipping (settings.json); the code does neither.
• Marks the unimplemented behavior as aspirational with a TODO pointing at follow-up kanban task 1e2c7.
• Implementation is intentionally out of scope — this is a Case B decision (per the discovery prompt: README is the spec, code is missing the feature; default to filing a follow-up rather than expanding this PR).

Trace evidence

Discovery during 2026-05-05 dogfood scan over sample_data/: 79 files scanned, 59 ingest / 20 skipped. coding_projects/my_website/ (3 files), coding_projects/rust_cli/ (2), coding_projects/data_analysis/ (3) all queued as personal_data. config/settings.json queued as personal_data too.

Code trace (read-only):

• src/ingestion/smart_folder/scanner.rs::scan_directory_recursive — only filters by extension (is_ingestible_file) and skips dotfiles (file_name.starts_with('.')). No manifest awareness anywhere in this file.
• src/ingestion/smart_folder/scanner.rs:489 — test_scan_includes_coding_project_files is an explicit test asserting the opposite of the README: coding-project files MUST be in file_paths. That's the contract the code defends today.
• src/ingestion/smart_folder/mod.rs::perform_smart_folder_scan_with_progress — orchestrates scan → image-dir classify → dedup → heuristic → LLM. No directory-skip step.
• src/ingestion/smart_folder/classify.rs::apply_heuristic_filtering — .json/.toml/.py/.rs/.js → should_ingest=true, category personal_data or code. Heuristic resolves coding-project files instantly, so they never reach the LLM where contextual judgment might have skipped them.
• .bashrc IS skipped — incidentally, by the scanner's dotfile filter, not by any config-aware rule. settings.json is NOT skipped — .json is heuristic-classified as personal data.

Why Case B (not Case A)

The discovery prompt's tiebreaker: "Default to Case B unless the spec is clearly broken — manifest-based auto-skip is a useful UX primitive for new users." It is — coding-project files dwarf personal data on most laptops, and the README's behavior is the right product. So spec stays, code gets the follow-up.

What this PR does NOT do

• No production code changes.
• No src/ingestion/smart_folder/* edits — the scanner's contract stays as-is until the follow-up task lands.
• No new tests — the scope is README-only.

Follow-up

Filed as kanban task 1e2c7 — feat(smart-folder): auto-skip directories with package.json / Cargo.toml / pyproject.toml manifests before LLM classification. The follow-up prompt includes the trace above plus an explicit plan to:

1. Add a manifest-list const next to is_ingestible_file.
2. Make scan_directory_recursive peek for the manifests before recursing; skip the entire subtree.
3. Replace test_scan_includes_coding_project_files with test_scan_skips_coding_project_directories.
4. Restore the README's "Auto-skip coding projects" wording (delete the TODO from this PR).

The settings.json drift is not part of that follow-up — it's an LLM-judgment problem rather than a manifest one. Flagged in the README; will need its own task if anyone wants it.

Test plan

• Manual code trace through scanner.rs, mod.rs, classify.rs confirming no auto-skip logic exists (covered above).
• Confirmed the existing test asserts the opposite of the README claim.
• Reviewed README diff — only sample_data/README.md changes; no Rust code touched.
• Re-running the dogfood scan would just re-confirm the discovery findings; not repeated here.

🤖 Generated with Claude Code