Commits on May 23, 2026

1. test: comprehensive coverage sweep + 2 real bug fixes uncovered by tests …

   97 → 177 tests this session (added 57 in this commit).
   
   ## Two real bugs caught by writing tests:
   
   1. **CLI rerun command crashed on import.**
      `clearscript projects rerun <slug>` imported TxtAdapter from
      `clearscript.ingest`, but that module only re-exports IngestAdapter
      and registry helpers — TxtAdapter lives in `clearscript.ingest.txt`.
      The web-API path works because it uses _FORMAT_ADAPTERS directly.
      Fixed by importing from the right submodule.
   
   2. **reject_term() never actually deprecated terms.**
      The schema defines a 'deprecated' status and `all_terms_in_domain`
      filters by `status != 'deprecated'`, but `reject_term` only updated
      reject_count + confidence — it never set status. So rejected terms
      kept resurfacing in subsequent library-context blocks. Fixed by
      flipping status to 'deprecated' on explicit reject.
   
   3. **rerun endpoint crashed when original provider was removed from config.**
      If a user ran v0.0.10 with provider X, then renamed/removed it,
      /api/projects/<slug>/rerun would 400 with "Provider 'X' not
      configured". Fixed by falling back to current default when the
      original's provider isn't in config anymore.
   
   ## New test files
   
   - **test_server.py** (33 tests) — TestClient-driven coverage of every
     HTTP endpoint: health/providers/example, /api/run sync + SSE stream,
     /api/run-file multipart, projects CRUD + transcript download,
     /api/projects/{slug}/rerun including provider-fallback case, library
     terms/speakers/patterns CRUD, accept-suggestions (Mode B), docx
     export, cost preview.
   - **test_provider_streaming.py** (5 tests) — verifies _BaseProvider's
     default chat_with_progress correctly wraps stream() into the
     (delta, payload)+ / (done, ChatResponse) protocol the pipeline expects.
   - **Extended test_cost.py** with 5 actual_cost() tests (the v0.0.9
     function had zero coverage).
   - **Extended test_cli.py** with 6 tests including end-to-end CLI rerun.
   - **Extended test_library.py** with 7 edge cases: lookup_alias for
     canonical, FTS partial matching, reject deprecation, delete cleans
     aliases, speaker re-add appends, pagination.
   - **Extended test_pleasantry_filter.py** with 7 _slug_hint_from_input
     tests covering title/filename/briefing/transcript priority order.
   
   All 177 tests pass. Ruff clean.

   

   Chen17-sq committed May 23, 2026
   Configuration menu

   • View commit details
   • Copy full SHA for 1de8623
   • Browse repository at this point

   Copy the full SHA

   1de8623 View commit details

   Browse the repository at this point in the history

2. test: parser edge cases + library updates + end-to-end compounding loop …

   Coverage push #2: 177 → 191 tests.
   
   ## New tests
   
   **tests/unit/test_pipeline.py** — 5 new:
   - _split_output handles model output with no delimiters (return raw)
   - _split_output survives malformed JSON in changelog section
   - _split_output filters non-dict entries inside the JSON arrays
   - _dedupe_suggestions collapses by kind+canonical (case-insensitive)
   - _dedupe_suggestions skips items with no identity
   
   **tests/unit/test_library.py** — 6 new:
   - update_term replaces (not appends to) the alias set
   - update_term changes domain end-to-end
   - update_speaker changes display label
   - Aliases are case-sensitive by design (lowercase variants don't match)
   - add_negative is idempotent on (text, do_not_change_to)
   - finish_session persists token counts
   
   **tests/integration/test_library_loop.py** — 3 new, the headline:
   - test_full_compounding_loop — proves the "library that compounds"
     claim end-to-end: run produces suggestion → accept-suggestions
     persists it → next run's system prompt actually contains the new
     mapping. If this test breaks, the product's central promise breaks.
   - test_rerun_uses_updated_library — re-running an old project against
     the freshly-updated library pulls the new mapping into the prompt.
   - test_seed_pack_terms_already_available_on_fresh_install — the 17
     seed terms (Dify/DeFi etc.) are usable on the very first run.
   
   All 191 tests pass. Ruff clean.

   

   Chen17-sq committed May 23, 2026
   Configuration menu

   • View commit details
   • Copy full SHA for 9d9484f
   • Browse repository at this point

   Copy the full SHA

   9d9484f View commit details

   Browse the repository at this point in the history

3. v0.0.12 — library as portable artifact + bulk ops + SDK-level tests …

   The library was trapped in a SQLite file at
   ~/Documents/clearscript/data/library/library.db. v0.0.12 makes it a
   first-class portable artifact, plus a sweep of UX polish and the
   provider-SDK test coverage gap that was lurking.
   
   ## What ships in this release
   
   1. **Library export / import** — both /api endpoints and CLI:
      - clearscript lib export <path>   → writes versioned JSON
      - clearscript lib import <path>   → merges (union of aliases)
      - Format marker so future versions detect incompatible files
      - Deprecated (rejected) terms excluded from export — sharing your
        library won't dump your rejections onto a teammate
   
   2. **CLI lib search** — FTS5 partial-match search to complement the
      existing exact-alias `lib lookup`
   
   3. **Bulk operations**
      - POST /api/library/terms/bulk-delete (cascades to aliases)
   
   4. **Rerun-of UI badge** — every project summary now exposes rerun_of;
      the web UI renders a ↻ rerun badge on cards with a tooltip pointing
      to the original slug. Provenance visible at a glance.
   
   5. **Real-SDK provider tests** (test_provider_sdks.py, 7 tests).
      AnthropicProvider and OpenAICompatProvider previously only got
      exercised via _BaseProvider fallbacks; the actual SDK call shapes
      were untested. Now covers:
      - Anthropic: messages.create kwargs, system extraction, multi-
        system concatenation, non-text content blocks, streaming with
        final-message usage capture
      - OpenAI-compat: chat.completions.create with include_usage in
        stream_options, fallback when usage missing
   
   ## Tests
   
   191 → 217. All passing. Ruff clean.
   
   Files: 9 new tests in test_library.py, 6 in test_server.py, 6 in
   test_cli.py, 7 in new test_provider_sdks.py.

   

   Chen17-sq committed May 23, 2026
   Configuration menu

   • View commit details
   • Copy full SHA for 35a43f0
   • Browse repository at this point

   Copy the full SHA

   35a43f0 View commit details

   Browse the repository at this point in the history