AutoGPT

mirror of https://github.com/Significant-Gravitas/AutoGPT.git synced 2026-04-08 03:00:28 -04:00

Author	SHA1	Message	Date
Zamil Majdy	553e224caf	Merge remote-tracking branch 'origin/dev' into feat/file-ref-structured-parsing-v2	2026-03-17 06:16:49 +07:00
Zamil Majdy	46acc2cfc1	fix(backend/copilot): update test mock targets after get_manager rename Tests were mocking the removed `get_manager` function. Updated mock targets to reference `get_workspace_manager` which replaced it.	2026-03-17 04:47:15 +07:00
Zamil Majdy	1c55c134ac	refactor(backend/copilot): reorder functions top-down and remove get_manager wrapper Address PR review comments from Pwuts: - Reorder file_ref.py: public API functions first, private helpers after - Reorder file_content_parser.py: public API first, parsers/helpers after - Remove get_manager() wrapper in workspace_files.py; callers now import get_workspace_manager directly from backend.copilot.context	2026-03-17 02:50:28 +07:00
Zamil Majdy	4f7bab3375	fix(backend/copilot): narrow sandbox exception handling, add missing tests, improve docs - Replace bare `except Exception` in sandbox file read with specific exception types (FileNotFoundError, OSError, UnicodeDecodeError) plus E2B SandboxException, re-raising unexpected exceptions as real bugs. - Expand __init__.py docstring explaining why lazy imports (PEP 562) are needed: tool_adapter uses TOOL_REGISTRY at module level, making the circular import cycle non-trivial to restructure. - Add docstring to _is_tabular explaining why isinstance checks are appropriate (structural type guard on opaque Any, not duck typing). - Add integration tests: bare ref binary format with line range (ignored), bare ref TOML parsing. - Add pyarrow skipif markers to file_ref_test.py binary format tests.	2026-03-17 00:30:38 +07:00
Zamil Majdy	981891aebe	fix(backend/copilot): address PR review comments - Narrow `except Exception` to specific types in `read_file_bytes` - Use `itertools.zip_longest` in `_tabular_to_list_of_dicts` - Add NaN handling test for parquet bare refs	2026-03-16 17:27:53 +07:00
Zamil Majdy	c44d8d7a98	fix(backend): address all remaining PR review items Blockers: - TOCTOU in local file read: replace os.path.getsize + open with a single read(_MAX_BARE_REF_BYTES + 1) to detect oversized files atomically - PARSE_EXCEPTIONS: replace filter(None,...) + type: ignore with _optional_exc() loader helpers that raise ImportError at parse time; removes all type: ignore suppressors - _check_content_size: raise ValueError instead of FileRefExpansionError so _expand_bare_ref can unify all resolution errors in one handler Should-fix: - _adapt_to_schema: non-tabular list to object-typed field now raises FileRefExpansionError instead of passing through silently; update test - _parse_delimited: extract _row_has_content predicate to module-level and use consistently in both the initial read and sniffer-fallback pass - _parse_xlsx: add engine="openpyxl" explicitly to pd.read_excel Nice-to-have: - sdk/__init__.py: convert sequential if branches to dispatch dict - _adapt_to_schema: extract _adapt_dict_to_array and _adapt_list_to_object helpers to stay under the 40-line limit - _parse_jsonl: cache keys_tuple = tuple(keys) to avoid O(n*k) allocations - _parse_yaml: add runtime warning when multi-document YAML (---) detected - is_media_file_ref (file.py): document URL-in-source-code false-positive risk	2026-03-16 06:19:54 +07:00
Zamil Majdy	1b7a446e97	test(backend): add 2 missing test cases from coderabbitai review - test_unknown_mime_falls_through_to_extension: verifies that when an unrecognized MIME is present, infer_format_from_uri falls through to extension-based detection (e.g. text/plain + .csv → csv) - test_bare_ref_binary_format_ignores_line_range: documents that binary bare refs (parquet/xlsx) silently drop the [start-end] range and parse full file content (line slicing is meaningless on binary bytes)	2026-03-15 22:58:01 +07:00
Zamil Majdy	d10ae1f392	fix(backend/copilot): narrow exception handling in _infer_format_from_workspace Replace bare `except Exception` catch-all with specific exceptions (ValueError, FileNotFoundError, OSError, PermissionError, AttributeError, TypeError) so unexpected programming errors (e.g. KeyError, RuntimeError) propagate instead of being silently swallowed as a None return. Addresses autogpt-reviewer comments #2930931556 and #2930341488.	2026-03-15 22:43:29 +07:00
Zamil Majdy	72b40f03bd	fix(backend): address remaining review comments — ArrowException, binary line range warning, CSV blank-line filter, _is_tabular refactor - Add pyarrow.ArrowException to PARSE_EXCEPTIONS so ArrowIOError and ArrowCapacityError (which don't inherit from ValueError/TypeError) are caught by the fallback contract in parse_file_content and _expand_bare_ref - Log a warning when a line range is specified for a binary format bare ref (parquet/xlsx) instead of silently ignoring it - Fix _parse_delimited blank-line filter: csv.reader never yields [] so 'if row' was a no-op; now uses 'if any(cell for cell in row)' to properly skip rows where every cell is empty - Refactor _is_tabular to early-return pattern, avoiding a double iteration over parsed[0] (once for isinstance(row, list) check, once for str check)	2026-03-15 22:24:43 +07:00
Zamil Majdy	575a956df2	fix(copilot): broaden exception handling in read_file_bytes for GCS errors The workspace branch of read_file_bytes only caught PermissionError and OSError, missing network/API exceptions from the gcloud.aio library (e.g. aiohttp.ClientError, asyncio.TimeoutError). Add a catch-all Exception handler that logs a warning and wraps the error in ValueError, consistent with the sandbox branch and _infer_format_from_workspace.	2026-03-15 14:05:18 +07:00
Zamil Majdy	6001f9dde0	fix(backend/copilot): optimize size guard with fast lower-bound check Add a fast lower-bound check in _check_content_size: if the character count already exceeds the byte limit, reject immediately without allocating an encoded copy. This avoids a ~10MB temporary allocation for strings near the size limit. The existing upper-bound fast path (char_len * 4 <= limit) is preserved for the common under-limit case. Addresses autogpt-reviewer should-fix item #3 from PR #12392.	2026-03-15 13:35:29 +07:00
Zamil Majdy	90d986973e	fix(backend): address autogpt-reviewer should-fix items for file parsing - Rename infer_format → infer_format_from_uri for clarity (all callers updated) - Add .xls format detection and helpful error message suggesting .xlsx conversion - Strengthen test_invalid_yaml_fallback to assert exact return value - Add tests: NaN in Parquet, header-only CSV, binary format + line range, .xls UX - Add docstring to _expand inner function in file_ref.py	2026-03-15 04:11:26 +07:00
Zamil Majdy	3638384f08	fix(backend): address PR review — integration tests and cleaner ragged row handling Add 6 integration tests for bare ref structured parsing in expand_file_refs_in_args (JSON, CSV, YAML, unknown extension, invalid JSON fallback, embedded ref stays string). Simplify _tabular_to_list_of_dicts to use zip + pad instead of index-based iteration, per reviewer suggestion.	2026-03-14 23:45:37 +07:00
Zamil Majdy	447a6f10fb	fix(backend): harden file parsing, add lazy imports, and improve file_ref safety checks	2026-03-14 23:00:46 +07:00
Zamil Majdy	c39e1a2701	fix: add workspace/sandbox size limits, deduplicate parse exceptions, case-insensitive MIME - Enforce 10MB size limit on workspace and sandbox file reads - Extract shared PARSE_EXCEPTIONS constant to eliminate duplication - Consolidate _create_blob/_create_binary_blob into single function - Add globals() caching for lazy PEP 562 imports - Fix case-insensitive MIME fragment matching per RFC 2045	2026-03-14 22:27:15 +07:00
Zamil Majdy	f037e02bd6	Merge remote-tracking branch 'origin/dev' into feat/file-ref-structured-parsing-v2	2026-03-14 22:11:09 +07:00
Zamil Majdy	4934f7a766	fix(backend/copilot): harden file-ref parsing — size guards, edge cases, sniffer fallback - Add pre-read size check in read_file_bytes for local files - Add OpenpyxlInvalidFile to parse_file_content exception list - Add csv.Sniffer fallback for misidentified delimiters - Add application/yaml MIME mapping - Fix _is_tabular to reject empty header rows - Fix _adapt_to_schema dict→List[str] not wrapping incorrectly - Fix _apply_line_range to note when range exceeds file - Remove inconsistent ChatSession string quoting - Add clarifying comments for budget/size check ordering - Add tests for all new behaviors	2026-03-14 21:54:52 +07:00
Zamil Majdy	eafac037c2	Merge remote-tracking branch 'origin/dev' into feat/file-ref-structured-parsing-v2	2026-03-14 10:21:00 +07:00
Zamil Majdy	5873dc81fe	fix(backend/copilot): catch openpyxl.InvalidFileException in bare ref parsing InvalidFileException extends Exception directly (not ValueError/OSError), so it wasn't caught by the existing exception tuple. Corrupt .xlsx files would crash instead of raising FileRefExpansionError.	2026-03-14 06:56:26 +07:00
Zamil Majdy	81c51f91f3	fix(backend): fix 2 CI test failures from media resolution changes - test_multi_file_commit_error_path: provide execution_context kwarg now required by GithubMultiFileCommitBlock.run() - test_non_media_string_field_still_reads_content: mock _infer_format_from_workspace to avoid event loop binding issue in CI	2026-03-14 06:44:23 +07:00
Zamil Majdy	1e03a56f28	fix(backend/blocks): address code review — type consistency, tests, docs - Use FileOperation enum instead of raw string in commits.py run() - Remove unused _MediaFileBlock test helper class - Add tests for resolve_media_content with data URI and HTTPS inputs - Clarify prompting docs: pass workspace:// directly, not via @@agptfile	2026-03-14 06:27:37 +07:00
Zamil Majdy	50b2dffc2e	fix(backend/blocks): preserve binary files in MediaFileType fields - Add `_is_media_file_field` check in file_ref.py `_expand` to pass through workspace:// URIs for `format: "file"` schema fields, skipping file reading and format inference that would corrupt binary content (images, videos, etc.) via UTF-8 decoding - Add `resolve_media_content` helper in file.py to encapsulate the is_media_file_ref + store_media_file pattern for blocks - Add `parse_data_uri` and `is_media_file_ref` utilities in file.py, consolidating regex from store_media_file and get_mime_type - Update GithubMultiFileCommitBlock to use resolve_media_content and parse_data_uri for proper binary blob encoding via GitHub API - Add LLM education in prompting.py about `format: "file"` fields and workspace:// references for media content - Add tests for all new functionality	2026-03-14 06:23:28 +07:00
Zamil Majdy	7b866a9f22	fix(backend): catch binary parser exceptions (BadZipFile, KeyError, TypeError) openpyxl.InvalidFileException and zipfile.BadZipFile extend Exception directly (not ValueError/OSError), so corrupt xlsx files would escape the narrowed exception handlers in both parse_file_content and _expand_bare_ref. Add KeyError (covers ArrowKeyError, OptionError), TypeError (covers ArrowTypeError), and zipfile.BadZipFile to both catch clauses.	2026-03-13 22:42:29 +07:00
Zamil Majdy	d48eb12d50	refactor(copilot): inline MIME format lookup in _infer_format_from_workspace Collapse 3-line MIME lookup (get → if → return fallback) into a single `return ... or ...` expression, as noted in self-review.	2026-03-13 22:05:20 +07:00
Zamil Majdy	ecbab7f3a0	refactor(backend): address PR review - extract helpers, DRY parsers, defer pandas - Extract `_to_str()` helper to deduplicate `content.decode("utf-8", errors="replace")` (3 call sites) - Extract `_check_content_size()` to consolidate duplicated size guard logic - Extract `_expand_bare_ref()` from `_expand` inner function (~90 lines -> ~30 lines each) - Move `expect_string` computation inside bare-ref branch (lazy evaluation) - Narrow `except Exception` in parse block to specific exception types - DRY `_parse_json`/`_parse_yaml` with `_parse_container` helper - Defer `import pandas` to binary parser functions to avoid ~200ms cold start - Tighten parser return type annotations	2026-03-13 19:54:11 +07:00
Zamil Majdy	b7551a1d18	fix(backend): use actual byte size for string size guard, narrow exception handling - Size guard in _expand now computes len(content.encode("utf-8")) for strings instead of len(content) which returns character count. This fixes a security issue where multi-byte UTF-8 strings (e.g. emoji) could pass up to 40MB through a 10MB byte limit. - Narrow except Exception in _infer_format_from_workspace to only catch expected IO/lookup failures (ValueError, FileNotFoundError, OSError, PermissionError). - Narrow except Exception in parse_file_content to only catch expected parse failures, letting programming bugs surface.	2026-03-13 19:40:20 +07:00
Zamil Majdy	f1366ea139	Merge remote-tracking branch 'origin/dev' into feat/file-ref-structured-parsing-v2	2026-03-13 19:27:04 +07:00
Zamil Majdy	68f430f792	fix(backend): restore PyYAML cp38 wheel hashes in poetry.lock Local poetry lock on Python 3.13 strips cp38 hashes that CI (Python 3.11) needs for the lock file consistency check.	2026-03-13 18:18:09 +07:00
Zamil Majdy	cb50c48eda	fix(backend): restore PyYAML cp38 hashes in poetry.lock and guard row truncation - Re-add PyYAML cp38 wheel hashes dropped by local poetry lock on Python 3.13 (CI needs them for Python 3.11) - Truncate rows to header length before dict conversion to prevent None keys when data rows exceed header column count	2026-03-13 18:17:22 +07:00
Zamil Majdy	b3af75a71e	Merge remote-tracking branch 'origin/dev' into feat/file-ref-structured-parsing-v2 # Conflicts: # autogpt_platform/backend/poetry.lock	2026-03-13 18:08:19 +07:00
Zamil Majdy	e0abb66ffc	refactor(backend): address reviewer comments on file ref PR - Narrow bare `except Exception` to `(PermissionError, OSError)` / `OSError` in read_file_bytes - Use itertools.zip_longest in _tabular_to_list_of_dicts for cleaner ragged-row handling - Tighten pyarrow version constraint from >=14.0.0 to ^23.0.0	2026-03-13 17:52:03 +07:00
Zamil Majdy	13985feb3d	fix(backend): extract list values from dict for List[List[Any]] block inputs When a dict (YAML/TOML) is passed to a List[List[Any]]-typed input (e.g. ConcatenateListsBlock), instead of wrapping the whole dict as [dict] (which causes pydantic to coerce dict → list of tuples), extract all list-typed values from the dict as inner lists. Example: YAML {"fruits": [{...},{...}]} → [[{...},{...}]] instead of [{"fruits": [...]}] which pydantic would coerce to [(key, val), ...]. Falls back to [dict] wrapping when no list values exist in the dict.	2026-03-13 16:40:41 +07:00
Zamil Majdy	bce0214a6a	fix(backend): pad ragged rows with None in tabular conversion helpers _tabular_to_column_dict and _tabular_to_list_of_dicts now use None for missing values in shorter rows instead of silently dropping them. This ensures all columns have equal length in column-dict output.	2026-03-13 16:26:54 +07:00
Zamil Majdy	a937433d8e	fix(backend): add isinstance(parsed, list) guards for pyright narrowing After isinstance(parsed, dict) on line 339, pyright narrows the type to Any \| dict and _is_tabular() alone doesn't narrow it back. Adding explicit isinstance(parsed, list) checks satisfies pyright's type narrowing.	2026-03-13 16:15:20 +07:00
Zamil Majdy	04bb0bef69	fix(backend): fix pyright type errors in tabular helper signatures Widen parameter type from list[list] to list so pyright doesn't complain about Any not being assignable to list[list] after _is_tabular() guard.	2026-03-13 16:04:06 +07:00
Zamil Majdy	e47ab6a295	fix(backend): convert tabular data to list-of-dicts for Any-typed block inputs FindInDictionaryBlock has `input: Any` which produces a JSON schema with no "type" key. _adapt_to_schema was only checking for explicit type matches ("object", "array"), so tabular data [[header],[rows]] passed through unchanged — causing FindInDict to return silent empty [] on key lookup. Fix: when prop_schema has no type (Any field), convert tabular data to list-of-dicts [{col: val, ...}, ...] which FindInDict's list-of-dicts branch (line 195-199) handles correctly. Refactored _adapt_to_schema to extract _is_tabular(), _tabular_to_list_of_dicts(), and _tabular_to_column_dict() helpers for clarity. Added 7 E2E tests covering CSV/TSV/JSONL/Parquet/YAML/JSON × Any-typed block.	2026-03-13 15:47:25 +07:00
Zamil Majdy	ad542db7fd	feat(backend): add schema-aware type adaptation for file ref expansion When parsed file content doesn't match the block's expected schema type, _adapt_to_schema() converts it to a more useful representation: - Dict → array schema: wraps dict in [dict] instead of pydantic flattening keys/values (fixes YAML/TOML dict → list block quirky coercion) - Tabular list → object schema: converts [[header], [rows]...] to column-dict {"col1": [vals], ...} (fixes CSV/JSONL/Excel → dict block empty results) Adds E2E tests for YAML→list, TOML→list, CSV→dict, JSONL→dict, and Parquet→dict adaptations.	2026-03-13 15:23:27 +07:00
Zamil Majdy	78cc963633	fix(backend): require ≥2 dicts for JSONL tabular detection Single-line JSONL with one dict was incorrectly converted to a table due to vacuously true all() on empty lines[1:]. Now requires ≥2 uniform dicts before applying table format conversion.	2026-03-13 15:03:47 +07:00
Zamil Majdy	d7b0e16112	feat(backend): JSONL tabular detection — uniform dicts → table format When all JSONL lines are dicts with identical keys (tabular data), _parse_jsonl now returns [header_row, data_row1, ...] instead of [{dict1}, {dict2}, ...]. This makes JSONL output consistent with CSV/TSV/Parquet/Excel and avoids awkward pydantic coercion when list[list] blocks receive a list of dicts. Heterogeneous JSONL (different keys or mixed types) keeps the existing behavior of returning a plain list of parsed values. Also adds comprehensive E2E tests for: - JSONL × block type matrix (list, string, dict) - Parquet × block type matrix (now that pyarrow is available)	2026-03-13 14:57:50 +07:00
Zamil Majdy	83e4a8a00d	poetry lock	2026-03-13 14:53:11 +07:00
Zamil Majdy	9fedd98a19	fix(backend): add pyarrow dependency for parquet file parsing pyarrow is required by pandas.read_parquet() but was never declared in pyproject.toml, causing parquet file parsing to fail at runtime.	2026-03-13 14:51:58 +07:00
Zamil Majdy	edc138fbec	fix(backend): reject binary formats (parquet/xlsx) passed to string-typed block inputs Binary formats decoded to UTF-8 produce garbled output. Instead of silently passing raw bytes as text, raise FileRefExpansionError with a clear message directing users to use structured (list/object) blocks.	2026-03-13 14:31:22 +07:00
Zamil Majdy	59278ba3de	fix(copilot): update integration test patches for get_manager → get_workspace_manager rename	2026-03-13 14:16:01 +07:00
Zamil Majdy	1fcd38ae6a	test(copilot): add e2e tests proving CSV→string block gets raw text after coercion Adds 3 end-to-end tests that exercise the full pipeline: expand_file_refs_in_args → coerce_inputs_to_schema - test_e2e_csv_to_string_block_no_json_dumps: proves the old bug (CSV parsed to list → json.dumps'd) is fixed — block gets raw CSV - test_e2e_csv_to_list_block_parses: proves non-string blocks still get structured parsed data - test_e2e_json_to_string_block_returns_raw_json: proves JSON file ref to string-typed block returns raw JSON text, not parsed dict	2026-03-13 14:03:38 +07:00
Zamil Majdy	c342ccc42c	fix(copilot): resolve circular imports and add full format×schema matrix tests - Move get_workspace_manager to copilot/context.py to break the tools → sdk → tools circular import chain - Make sdk/__init__.py use PEP 562 lazy loading to avoid eagerly importing service.py (which chains to tools/) - Replace local import in run_block.py with top-level import - Add parametrized matrix tests covering JSON/CSV/TSV/JSONL/YAML/TOML × string-typed/non-string/opaque-object schema combinations - Add two-phase expansion integration test (tool-level skip + block-level expand)	2026-03-13 13:56:03 +07:00
Zamil Majdy	79b881393e	fix(backend): surface actual parse errors for binary formats and skip parsing for string-typed inputs - Add strict mode to parse_file_content that propagates exceptions instead of silently falling back, used for binary formats (parquet/xlsx) so the actual error (missing library, corrupt file) is shown instead of a blanket message - Make expand_file_refs_in_args schema-aware: when the tool input schema declares a parameter as type "string", structured parsing is skipped and raw file content is returned as-is (e.g., CSV text stays as CSV text, not json.dumps of parsed rows) - Pass input_schema from tool_adapter through to file ref expansion	2026-03-13 13:28:30 +07:00
Zamil Majdy	9300b3223f	fix(backend): resolve workspace file format from metadata when URI has no extension Workspace URIs by ID (workspace://abc123) have no file extension and may lack a MIME fragment. When infer_format returns None, fall back to querying the workspace file manager for the file's stored MIME type and original filename to determine the correct parser. Also rename _MIME_TO_FORMAT → MIME_TO_FORMAT (public) for reuse.	2026-03-13 12:45:18 +07:00
Zamil Majdy	d5136891bf	fix(backend): raise clear error when binary format parsing fails When parse_file_content falls back for binary formats (parquet/xlsx), detect the fallback and raise FileRefExpansionError with a clear message about missing dependencies instead of silently returning garbled UTF-8 decoded bytes.	2026-03-13 05:41:29 +07:00
Zamil Majdy	3b4586bbe6	fix(backend): guard pd.isna() against non-scalar cells in _df_to_rows pd.isna() on a list/dict returns a boolean array which raises ValueError in a boolean context. Add _is_nan() helper that checks is_scalar first.	2026-03-13 04:19:55 +07:00
Zamil Majdy	bc1d68310c	fix(backend): fix pyright errors in file_content_parser_test - Add pyright: ignore[reportMissingImports] for optional pyarrow import - Add type: ignore[arg-type] for BytesIO passed to to_excel	2026-03-13 04:18:26 +07:00
Zamil Majdy	abd59d4d1d	fix(backend): fix NaN→None normalization in _df_to_rows for float64 columns df.where(df.notna(), None) silently converts None back to NaN in float64 columns. Use explicit pd.isna() cell-level checking instead. Add test to verify no NaN leaks into serialized output.	2026-03-13 03:56:17 +07:00
Zamil Majdy	d87d522706	feat(backend): add bare-ref integration tests and fix size guard allocation - Add 6 integration tests for bare-ref structured parsing in file_ref_test.py (JSON, CSV, unknown ext, invalid JSON, embedded ref, oversized) - Remove unnecessary .encode("utf-8") allocation in size guard — use len() as a lower-bound proxy for character count - Add per-ref truncation for unknown-format bare refs to _MAX_EXPAND_CHARS - Normalize bytes fallback to string for binary format parse failures	2026-03-13 03:54:07 +07:00
Zamil Majdy	262c1811b5	fix(backend): address Sentry review — truncate unknown-format bare refs and normalize bytes fallback - Apply _MAX_EXPAND_CHARS (200k) truncation to bare refs with unknown format, matching the inline expansion behavior - Normalize bytes→str when parse_file_content falls back to raw bytes for binary formats, ensuring tools never receive raw bytes	2026-03-13 03:49:43 +07:00
Zamil Majdy	48d4f3716b	fix(backend): skip parquet tests when pyarrow is not installed pyarrow is not bundled as a direct dependency — the fallback contract handles the missing engine at runtime. Skip the fixture-based test that needs pyarrow to generate test data.	2026-03-13 03:45:03 +07:00
Zamil Majdy	340b18d40b	fix(backend): address PR review comments for file content parser - Move local imports (yaml, pandas) to top-level - Add size limit guard (_MAX_BARE_REF_BYTES=10MB) for bare ref expansion - Ignore line ranges for binary formats (parquet/xlsx) instead of garbling - Strip query strings in infer_format URI parsing - Fix type annotations for parser dicts (Callable instead of Any) - Replace NaN with None in parquet/xlsx output for JSON serializability - Fix TOML empty dict fallback (always return parsed dict) - Remove .xls/.vnd.ms-excel mappings (openpyxl doesn't support legacy .xls) - Add posixpath.splitext usage comment - Update prompt to mention first-sheet-only and .xls limitation - Restore pyyaml cp38 entries in poetry.lock	2026-03-13 03:28:31 +07:00
Zamil Majdy	35f54549b5	feat(backend/copilot): parse @@agptfile bare refs by file extension When the entire tool argument is a bare @@agptfile: reference, the resolved content is now parsed based on the file's extension or MIME type instead of content-sniffing. Supported formats: JSON, JSONL, CSV, TSV, YAML, TOML, Parquet, Excel. Unrecognised formats or parse failures fall back to plain string.	2026-03-13 02:58:00 +07:00

Compare commits

56 Commits

dev ... feat/file-

Diff Content Not Available

Compare commits

56 Commits dev ... feat/file-

Diff Content Not Available

56 Commits

dev ... feat/file-