@djordje-stojanovic/sigmaskills 0.16.4 → 0.16.6

package/CHANGELOG.md

@@ -4,6 +4,49 @@ All notable changes to [Sigmaskills](https://github.com/Djordje-Stojanovic/Sigma
 
 ## [Unreleased]
 
+## [0.16.6] — 2026-10-23
+
+### Fixed
+
+- `sigmadeeplook` fix round 11, lane F1 (single-record lookups):
+  - New `Find one record:` line in step 1. It names the table, the row count, the unique key column with its first and last key, and the route (`--ask` or `--sql`). It prints only when a unique key exists. Covered: delimited text, `.po`/`.mo`, `.properties`, `.sbv`/`.srt`/`.vtt`, SAS, SPSS, Stata, netCDF (the dimension names), sqlite/gpkg, xlsx/xlsb/ods, jsonl and Parquet.
+  - `--ask` returns one record by its key: `Answer: <column> = <value> (where <key column> = <key>; <unit> <position>)`, then the record in full. Keys match across case, accents, Cyrillic or Latin letters and hyphens. A bare number matches exactly. A number after a column name (`visit_id 5650000`, `cue number 840`) is a filter.
+  - `--ask` now reads `.sav` (not ZLIB-compressed), SAS and Stata files, `.po`/`.mo` messages, subtitle cues and `.properties` keys as tables. A table over 1,000,000 cells that is asked a pure lookup is read in one pass and only the rows that hold a value of the question are kept (`SIGMADEEPLOOK_ASK_STREAM_CAP`).
+  - `--ask` reads one netCDF cell by index or coordinate value (`last time index and station index 17`, 2 decimals), unpacks `scale_factor`/`add_offset` and names fill values.
+  - A comparison (`Petal_Length greater than 5.0`) now uses the column whose whole name stands before it. A table inside a text keeps original file line numbers.
+- `sigmadeeplook` fix round 11, lane F3 (set 10 miss causes: structure and counts):
+  - The identity line shows the file's first 8 bytes in hex, in file order (`magic 950412de 00000000`), for every kind found by its magic bytes.
+  - Key/value, YAML and TOML files print a `Comment lines` fact: how many comment lines, at which line numbers, and the first comment block as written. A file version stated in a comment is now in step 1.
+  - `--ask "how many times does the word 'X' appear"` and `"in how many paragraphs ..."` count whole words over the full text of a Word, OpenDocument, PDF, PowerPoint or RTF file, case-sensitive when the quoted word has a capital, and name the text they counted. Before, the agent counted over a clipped print and was off by 1 to 4. A document look also lists its ten most frequent capitalised words.
+  - Office packages (`.docx`, `.xlsx`, `.pptx`, `.odt`, `.ods`, `.odp`) print their hyperlinks with targets, the pictures and media files with path and size, and footnote, endnote and comment counts.
+  - Container headers are read from the file's own bytes, with or without 7-Zip: a cabinet's folders (compression groups), files, data blocks and files per folder; the zip end record; the 7z start header; RAR version and flags; the ISO volume; the OLE sector layout.
+  - New: `.rtf` files are read (text, paragraphs, properties, table rows, picture and footnote counts).
+  - A `.docx` with a damaged part is flagged `CORRUPT` instead of `INTERNAL-ERROR`.
+- `sigmadeeplook` fix round 11, lane F2 (set 10 miss causes):
+  - One pixel of a picture is a lookup: `--ask "RGB colour at pixel x=520, y=600"` answers R,G,B, hex, alpha and palette index, counted from the top-left corner as displayed (bottom-up BMP, palettes, RLE, 24-bit PCX planes, EXIF turns). A pixel outside the picture is answered with the size. The picture print says where to ask (`Ask a pixel`) and gives the size. TGA and Netpbm files are now detected as pictures.
+  - A recording with 2 or more channels prints tones per channel (LEFT, RIGHT, channel 3 ...): count, start times, length and background level per frequency. Short pips of 25 ms and up are found. This works for .au, WAV (with or without ffmpeg) and any audio ffmpeg decodes. Before, only the first channel was read.
+  - A container with several streams prints a stream table: the container id (PID of a transport stream, track id of an MP4), type, codec, language, channels and bit rate. Every audio stream prints its own steady pitch. Without ffprobe a transport stream's PAT and PMT are read in pure Python, so the PIDs, stream types, languages and the service name are still printed.
+
+## [0.16.5] — 2026-10-23
+
+### Fixed
+
+- `sigmadeeplook` fix round 10, lane F1 (set 9 miss causes):
+  - A table inside a text is read from its own header row. A weighing protocol with a title block and a totals block was read as a one-column CSV with 100 reject rows. The title and total lines are now printed whole, and the last 20 rows of every table are printed as the file has them.
+  - A legacy Word file with junk in the high half of a stream size lost its text (`CORRUPT: no WordDocument stream`). Version 3 OLE files now ignore those bits, as the format requires.
+  - `.ppt` and `.xls` files print their saved properties: application, revision, dates, word and slide counts. Small `.doc` and `.odt` tables print every row. `.odt` footnote and endnote text is printed and kept in `text.md`.
+  - A last-record or fixed-position value with no blank in it (a URL, path or key) is no longer cut. A cut feed URL had given a confident wrong answer.
+  - New counts: `.po` flag lines per flag; `.properties` and INI keys per prefix; items under each node of a self-nesting XML tree (OPML); each distinct `.sbv`/subtitle line with its cue count when the text repeats.
+  - `.properties` values are read as Java reads them: escapes decoded, continued lines joined, escaped separators kept in the key. Latin-1 Spanish (`¿`, `ñ`) is no longer read as Windows-1250.
+- `sigmadeeplook` fix round 10, lane F2 (set 9 miss causes):
+  - SAS `.sas7bdat` files print their header: dataset name, type, dates, SAS release and host. The header is read without pandas, so a cut-off file still names itself, and its unreadable rows are flagged `CORRUPT`.
+  - Every worksheet table (`.xlsx`, `.xlsb`, `.ods`) prints its last row, hidden sheets too. A CSV inside an archive, nested ones too, prints its first and last row.
+  - netCDF: a packed variable prints its unpacked values (value x `scale_factor` + `add_offset`) and a grid of the last time step at round indexes, with the coordinate values. Char variables print every entry. A lone record variable is read without padding, and a damaged record count stops at the end of the file.
+  - Pictures print every palette entry, the box each frequent colour fills, and a colour gradient as a formula when one fits (for example `R = 255·x/1599, rounded down`). This works with or without numpy.
+  - Sound prints tones: stretches where one frequency holds most of the power, with their count, frequency and start times per frequency. A 40 MB file takes under a second.
+  - Video prints its key frames from the packet flags: count, first and last time, spacing.
+  - A video look with `--quick` printed nothing. Two threads swapped the standard output at once while they imported packages. Imports now take turns.
+
 ## [0.16.4] — 2026-10-23
 
 ### Fixed
@@ -438,7 +481,9 @@ Ship high-rigor, installable agent skills (not apps): full-repo audits, performa
 | 2026-08-11 | `5dab832` | Add CHANGELOG and mark v0.1.0 as the first Sigmaskills release |
 | 2026-08-11 | `124a666` | Polish v0.1.0 docs for multi-host install and use |
 
-[Unreleased]: https://github.com/Djordje-Stojanovic/Sigmaskills/compare/v0.16.4...HEAD
+[Unreleased]: https://github.com/Djordje-Stojanovic/Sigmaskills/compare/v0.16.6...HEAD
+[0.16.6]: https://github.com/Djordje-Stojanovic/Sigmaskills/compare/v0.16.5...v0.16.6
+[0.16.5]: https://github.com/Djordje-Stojanovic/Sigmaskills/compare/v0.16.4...v0.16.5
 [0.16.4]: https://github.com/Djordje-Stojanovic/Sigmaskills/compare/v0.16.3...v0.16.4
 [0.16.3]: https://github.com/Djordje-Stojanovic/Sigmaskills/compare/v0.16.2...v0.16.3
 [0.16.2]: https://github.com/Djordje-Stojanovic/Sigmaskills/compare/v0.16.1...v0.16.2

package/README.md

@@ -6,7 +6,7 @@ Portable [Agent Skills](https://agentskills.io/) that install once and run on **
 
 | | |
 |---|---|
-| **Release** | [**v0.16.4**](https://github.com/Djordje-Stojanovic/Sigmaskills/releases/tag/v0.16.4) |
+| **Release** | [**v0.16.6**](https://github.com/Djordje-Stojanovic/Sigmaskills/releases/tag/v0.16.6) |
 | **Changelog** | [CHANGELOG.md](CHANGELOG.md) |
 | **License** | [MIT](LICENSE) |
 | **Spec** | [agentskills.io](https://agentskills.io/) |

package/manifest.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
   "name": "sigmaskills",
-  "version": "0.16.4",
+  "version": "0.16.6",
   "description": "Portable Agent Skills monorepo and first-party Sigma Installer",
   "skills": [
     {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@djordje-stojanovic/sigmaskills",
-  "version": "0.16.4",
+  "version": "0.16.6",
   "description": "Portable Agent Skills monorepo and first-party Sigma Installer",
   "type": "module",
   "license": "MIT",