{ "specVersion": "1.0", "host": { "displayName": "StealthStack", "identifier": "did:web:stealthstack.ai" }, "entries": [ { "identifier": "urn:air:stealthstack.ai:skill:archives", "displayName": "Archive Container Reader & Converter", "type": "application/ai-skill+md", "url": "https://stealthstack.ai/skills/archives/SKILL.md", "trustManifest": { "identity": "did:web:stealthstack.ai", "identityType": "did" }, "description": "Read, list, and extract archive containers (.zip, .tar, .gz, .7z) to JSON or text. Use when inspecting or unpacking compressed file bundles.", "tags": [ "archive", "zip", "tar", "gzip", "7z", "extraction", "compression", "conversion" ], "capabilities": [ "ListContents", "ExtractFiles", "ReadArchive", "ConvertToJson", "InspectMetadata" ], "representativeQueries": [ "Read a .zip file and show me what's inside", "Extract files from a tar.gz archive", "List contents of a .7z archive", "Convert archive contents to JSON", "Inspect files in a compressed archive" ], "version": "0.1.0", "updatedAt": "2026-06-20T03:53:41.203Z", "metadata": { "context": "- .zip: local file headers at byte offsets + central directory at end-of-file. Python stdlib zipfile handles deflate, stored, bzip2, lzma. Password-protected ZIPs are not supported by this script (no --password argument); the stdlib API does accept a pwd= parameter but it is not wired up here.\n- .tar / .tar.gz / .tar.bz2 / .tar.xz: GNU/POSIX tape archive, optionally compressed. Use tarfile.open() with mode 'r:*' to auto-detect compression. tarfile does NOT support zip64 extra headers.\n- .gz (bare gzip, not tar): a single compressed file — gzip wraps exactly one stream. Use gzip.open() to decompress. There is no file listing; the original filename is stored in the header (fname field) but is optional.\n- .7z: proprietary format. The Python stdlib has NO native 7z support. The py7zr third-party library is required for extraction. Without it, the script reports the magic bytes (37 7A BC AF 27 1C) and exits with a clear message. The script uses subprocess to call the system `7z` CLI if available as a fallback.\n- ZIP path traversal: zipfile raises BadZipFile for structurally invalid ZIPs; always sanitize extracted paths (strip leading '/' and '..') when writing to disk.\n- tarfile path traversal: the script manually iterates members with extractfile() rather than calling tarfile.extractall(), so the filter parameter is not used. Path traversal is prevented by sanitizing member paths before writing to disk.\n- Symlink bombs: tar archives can contain symlinks pointing outside the destination. The script skips non-regular-file members (symlinks, devices) during --extract-all.\n- Large archives: listing does not decompress member content — it reads only headers. Extraction can be selective (single member) to avoid expanding multi-GB archives.\n- .tar.gz is the same as .tgz; tarfile.open() with 'r:*' handles both extensions identically.", "context_verified": true } }, { "identifier": "urn:air:stealthstack.ai:skill:ard-skill-builder", "displayName": "ARD Skill Builder", "type": "application/ai-skill+md", "url": "https://stealthstack.ai/skills/ard-skill-builder/SKILL.md", "trustManifest": { "identity": "did:web:stealthstack.ai", "identityType": "did" }, "description": "Author StealthStack ARD-1 SKILL.md files that drop straight into the catalog. Use when creating, scaffolding, or validating a new skill for the ARD registry.", "tags": [ "meta", "skill-authoring", "ard", "catalog", "builder" ], "capabilities": [ "ClassifyPrimitive", "ComposeSkillMd", "GenerateRepresentativeQueries", "ExtractContext", "ValidateFrontmatter" ], "representativeQueries": [ "Create a new ARD skill for sending Slack messages", "Turn this API into a catalog skill", "Scaffold a SKILL.md that passes ARD validation", "Help me author a skill for the StealthStack catalog", "Generate a skill folder I can drop into ard-1/public/skills" ], "version": "0.1.0", "updatedAt": "2026-06-19T20:56:57.000Z", "metadata": { "context": "- The catalog generator maps frontmatter -> one ARD entry of type application/ai-skill+md; never set auto fields (identifier, type, url, trustManifest, updatedAt, host).\n- representativeQueries MUST be 2-5 items or the field is dropped (losing the ranking surface) — aim for 4-5 real user phrasings, not keywords.\n- The folder name IS the slug, URL and URN; it must equal frontmatter `name` (kebab-case), be unique and stable. A folder starting with `_` is an excluded draft.\n- `context` is the trust root: extract from real source only, never invent; ship `context_verified: false` until a human signs off. (Unlike representativeQueries, do not free-generate it.)\n- Faithful-minimal beats example-rich: fabricated tool slugs/params/examples are the top failure mode. Don't claim capabilities the skill can't deliver.", "context_verified": false } }, { "identifier": "urn:air:stealthstack.ai:skill:arrow", "displayName": "Arrow / Feather Reader & Converter", "type": "application/ai-skill+md", "url": "https://stealthstack.ai/skills/arrow/SKILL.md", "trustManifest": { "identity": "did:web:stealthstack.ai", "identityType": "did" }, "description": "Read, inspect, and convert Apache Arrow columnar files (.arrow, .feather) to CSV or JSON. Use when opening or transforming Arrow/Feather data files.", "tags": [ "data", "arrow", "feather", "columnar", "csv", "json", "converter", "extractor" ], "capabilities": [ "ReadArrow", "InspectSchema", "ConvertToCsv", "ConvertToJson", "ExtractSample" ], "representativeQueries": [ "Read a .arrow or .feather file and show me the data", "Convert an Arrow file to CSV", "What is the schema of this .feather file?", "Extract a sample of rows from an Arrow dataset", "Convert Arrow columnar data to JSON" ], "version": "0.1.0", "updatedAt": "2026-06-20T03:53:41.204Z", "metadata": { "context": "- Apache Arrow IPC format uses two sub-formats: IPC Stream (no magic bytes, unbounded record batches) and IPC File/Random-Access (magic \"ARROW1\" at offset 0, allows seeking). The script tries ipc.open_file first and falls back to ipc.open_stream on failure — pyarrow itself does not auto-detect the sub-format.\n- Feather v1 (legacy) and Feather v2 (Arrow IPC File with LZ4/ZSTD compression) are distinct on-disk formats. pyarrow.feather.read_feather handles both transparently.\n- Column types map to Arrow types (int8..int64, float16/32/64, utf8, large_utf8, binary, list, struct, dictionary). Dictionary-encoded columns decode to their value types on to_pandas().\n- Null handling: Arrow distinguishes null (missing) from NaN (float not-a-number). The converter uses csv.writer directly (not pandas): Arrow nulls become empty CSV fields; NaN floats appear as the string 'nan'.\n- Large files: reading the full table into memory can OOM. Use pyarrow.dataset or ipc.open_file(...).get_batch(i) to page through record batches. For JSON output the converter buffers all records in memory before writing; use --rows N to cap output size.\n- pyarrow requires: pip install pyarrow (includes both arrow and feather support).", "context_verified": true } }, { "identifier": "urn:air:stealthstack.ai:skill:avro", "displayName": "Avro Reader & Converter", "type": "application/ai-skill+md", "url": "https://stealthstack.ai/skills/avro/SKILL.md", "trustManifest": { "identity": "did:web:stealthstack.ai", "identityType": "did" }, "description": "Read, inspect, and convert Apache Avro (.avro) files to CSV, JSON, or JSONL. Use when opening .avro files, extracting embedded schema, or exporting records.", "tags": [ "data", "avro", "schema", "csv", "json", "conversion", "extractor" ], "capabilities": [ "ReadAvro", "InspectSchema", "ConvertToCsv", "ConvertToJson", "ConvertToJsonl", "SampleRows" ], "representativeQueries": [ "Read a .avro file and show me the data", "Convert avro to CSV", "Extract the schema from an avro file", "Convert avro to JSON or JSONL", "Sample the first rows of an avro file" ], "version": "0.1.0", "updatedAt": "2026-06-20T03:53:41.204Z", "metadata": { "context": "- Avro is a binary row-based format; it cannot be read as plain text — always use fastavro or a compatible library.\n- Every Avro file embeds its writer schema in the file header (the first object container block). The schema is always recoverable from the file itself without any external registry.\n- Avro supports rich types: record, array, map, union, enum, fixed, bytes, and logical types (date, time-millis, timestamp-millis, decimal). Union fields commonly appear as [\"null\", \"someType\"] for nullable columns — fastavro resolves unions to Python None or the concrete value automatically.\n- Logical type \"decimal\" is stored as bytes/fixed with scale/precision metadata; fastavro returns it as a Python Decimal — cast to string or float before JSON serialisation.\n- Large Avro files are written in sync-delimited blocks; fastavro iterates records lazily (reader is a generator) so memory usage is low even for multi-GB files.\n- fastavro's `reader` returns dicts by default; nested records become nested dicts, arrays become lists, maps become dicts.\n- Avro does NOT store column statistics or row counts in the header — you must iterate all records to count rows.\n- Schema evolution (reader vs writer schema) is supported: pass a reader_schema to fastavro.reader to project/promote fields.", "context_verified": true } }, { "identifier": "urn:air:stealthstack.ai:skill:dbf", "displayName": "dBase DBF Reader & Converter", "type": "application/ai-skill+md", "url": "https://stealthstack.ai/skills/dbf/SKILL.md", "trustManifest": { "identity": "did:web:stealthstack.ai", "identityType": "did" }, "description": "Reads and converts dBase .dbf binary table files to text, CSV, or JSON. Use when inspecting or extracting data from legacy GIS, FoxPro, or dBase databases.", "tags": [ "dbf", "dbase", "foxpro", "extractor", "csv", "json", "converter", "gis", "legacy" ], "capabilities": [ "ReadDbf", "ConvertToCsv", "ConvertToJson", "DumpSchema", "InspectFields" ], "representativeQueries": [ "read a .dbf file and show me what's inside", "convert dbf to CSV", "extract data from a dBase or FoxPro database file", "what fields are in this .dbf file", "export a DBF table to JSON" ], "version": "0.1.0", "updatedAt": "2026-06-20T03:53:41.206Z", "metadata": { "context": "- dBase .dbf files begin with a 1-byte version code (0x03 = dBase III+, 0x83 = with memo, 0x30 = Visual FoxPro, 0xF5 = FoxPro 2.x with memo). Files that don't start with a recognised version byte are likely not valid .dbf files.\n- The header encodes number of records (bytes 4-7, little-endian uint32), header size (bytes 8-9), and record size (bytes 10-11). Field descriptors start at byte 32, each 32 bytes long, terminated by a 0x0D byte.\n- Field types: C (Character), N (Numeric), F (Float), D (Date, YYYYMMDD), L (Logical, T/F/Y/N), M (Memo pointer — needs a .dbt or .fpt sidecar), I (Integer, FoxPro), B (Double, FoxPro), @ (DateTime, FoxPro), + (Autoincrement, FoxPro).\n- Character encoding is not stored in the header in older versions; byte 29 of the header is the \"language driver ID\" (codepage mark) in dBase IV+ and FoxPro. Common values: 0x00 = unknown/DOS 437, 0x03 = Windows ANSI (CP1252), 0x04 = Macintosh, 0x64 = Eastern European (CP1250). When unknown, try cp1252 then fall back to latin-1.\n- The `dbfread` Python library (pip install dbfread) handles version detection, codepage lookup, and memo sidecar files automatically. Use it over raw struct parsing unless dbfread fails.\n- Deleted records have a '*' (0x2A) in their first byte; dbfread skips them by default (set `deleted=True` to include). Raw struct parsers must skip 0x1A (EOF marker) records.\n- Memo fields (.dbt for dBase III/IV, .fpt for FoxPro/Visual FoxPro) must reside in the same directory as the .dbf. Missing sidecar = memo values read as None/empty.\n- Numeric fields stored as blank strings represent NULL; do not coerce to 0.\n- Date fields come back as datetime.date objects from dbfread; serialise via .isoformat() for JSON.", "context_verified": true } }, { "identifier": "urn:air:stealthstack.ai:skill:doc", "displayName": "Legacy Word (.doc) Reader & Converter", "type": "application/ai-skill+md", "url": "https://stealthstack.ai/skills/doc/SKILL.md", "trustManifest": { "identity": "did:web:stealthstack.ai", "identityType": "did" }, "description": "Read and convert legacy OLE/CFB binary Word .doc files to plain text, JSON, or CSV. Use when you need to open or extract content from pre-2007 Word documents.", "tags": [ "document", "word", "doc", "ole", "cfb", "convert", "extract", "legacy" ], "capabilities": [ "ReadDoc", "ConvertToText", "ConvertToJson", "ConvertToCsv", "ExtractMetadata" ], "representativeQueries": [ "Read a .doc file and show me the text", "Convert a legacy Word doc to plain text", "Extract content from a .doc file to JSON or CSV", "Open an old Word document and get the text out", "Convert a pre-2007 .doc file to a readable format" ], "version": "0.1.0", "updatedAt": "2026-06-20T03:53:41.206Z", "metadata": { "context": "- .doc files use the OLE Compound File Binary (CFB) format, not ZIP/XML like .docx. Standard ZIP tools and XML parsers will fail on them.\n- The python-docx library only supports .docx (OOXML); it cannot open .doc files at all.\n- Recommended extraction libraries: olefile (low-level OLE stream reader, `pip install olefile`), textract, or antiword CLI. Note: olefile and oletools (python-oletools) are separate packages with different purposes — olefile reads OLE streams, oletools is a security-analysis toolkit. antiword is fast but may not be available on all platforms. textract wraps several backends (antiword, LibreOffice) and is the most portable Python option.\n- LibreOffice --headless can batch-convert .doc to .docx or .txt: `soffice --headless --convert-to txt `. Reliable but slow (LibreOffice runtime initialisation overhead, not JVM — LibreOffice is a C++ application).\n- Binary .doc files may contain embedded OLE objects, macros (VBA), revision marks, and proprietary fields that lose fidelity on extraction. Track-changes and comments are often lost.\n- Password-protected .doc files will fail extraction silently or with a generic error; surface this clearly to the user.\n- Encoding issues are common: older .doc files may use Windows-1252 or other legacy codepages. Use errors='replace' or errors='ignore' when decoding extracted bytes.", "context_verified": true } }, { "identifier": "urn:air:stealthstack.ai:skill:eml", "displayName": "EML Email Reader & Converter", "type": "application/ai-skill+md", "url": "https://stealthstack.ai/skills/eml/SKILL.md", "trustManifest": { "identity": "did:web:stealthstack.ai", "identityType": "did" }, "description": "Reads and converts RFC 822 email files (.eml) to plain text, JSON, or CSV. Use when opening .eml files, extracting email content, headers, or attachments metadata.", "tags": [ "eml", "email", "rfc822", "document", "converter", "mime" ], "capabilities": [ "ReadEml", "ConvertToJson", "ConvertToCsv", "ConvertToText", "ExtractHeaders", "ExtractAttachmentMetadata" ], "representativeQueries": [ "Read a .eml file and show me the email content", "Convert an eml email to JSON or CSV", "Extract headers from an .eml file", "What attachments are in this email file?", "Parse an RFC 822 email message" ], "version": "0.1.0", "updatedAt": "2026-06-20T03:53:41.206Z", "metadata": { "context": "- EML files follow RFC 822 / RFC 2822 / RFC 5322. They are plain-text files with headers (From, To, Subject, Date, Message-ID, etc.) separated from the body by a blank line.\n- Multi-part messages use MIME (RFC 2045-2049): the Content-Type header includes a boundary string that delimits each part. A single email may nest multipart/mixed, multipart/alternative, and multipart/related parts.\n- The body part to display to humans is typically the last text/plain or text/html leaf in a multipart/alternative block; prefer text/plain over text/html for readability.\n- Header values may be encoded as RFC 2047 encoded-words: =?charset?encoding?text?= (B=base64, Q=quoted-printable). Python's email.header.decode_header() handles this.\n- Attachment parts have Content-Disposition: attachment (sometimes inline) and a filename parameter. Payload is usually base64-encoded; email.message.get_payload(decode=True) decodes it.\n- Dates in the Date header follow RFC 2822 format and are returned as decoded strings; email.utils.parsedate_to_datetime() can be used downstream to convert them to datetime objects if needed.\n- Some EML files omit the leading \"From \" (mbox separator). The script reads raw bytes, decodes to a string (UTF-8, then latin-1 fallback), and passes the result to email.message_from_string().\n- Charset defaults to ASCII/US-ASCII when not declared; real-world emails often include UTF-8 or ISO-8859-1 without declaring it — decode with errors='replace' to avoid crashes.", "context_verified": true } }, { "identifier": "urn:air:stealthstack.ai:skill:epub", "displayName": "EPUB Reader & Converter", "type": "application/ai-skill+md", "url": "https://stealthstack.ai/skills/epub/SKILL.md", "trustManifest": { "identity": "did:web:stealthstack.ai", "identityType": "did" }, "description": "Reads and converts EPUB e-books to plain text, JSON, or CSV. Use when opening .epub files, extracting book content, or converting e-books to structured data.", "tags": [ "epub", "ebook", "document", "converter", "text-extraction" ], "capabilities": [ "ReadEpub", "ConvertToJson", "ConvertToCsv", "ConvertToText", "ExtractMetadata" ], "representativeQueries": [ "Read a .epub file and show me the content", "Convert an epub to JSON or CSV", "Extract text from an epub e-book", "Get the chapter list from an epub", "What metadata is in this epub file?" ], "version": "0.1.0", "updatedAt": "2026-06-20T03:53:41.207Z", "metadata": { "context": "- An EPUB is a ZIP archive. The container file at META-INF/container.xml points to the OPF package document (usually content.opf or similar), which lists spine items (reading order) and metadata.\n- Spine items reference XHTML/HTML documents inside the archive; reading order follows elements, not the filesystem order.\n- Metadata fields live in the OPF block: dc:title, dc:creator, dc:language, dc:publisher, dc:date, dc:identifier, dc:description.\n- Body text is inside XHTML files; strip tags to get plain text. Images, CSS, and fonts are also bundled but contain no readable prose.\n- EPUB 2 uses NCX (toc.ncx) for the table of contents; EPUB 3 uses a nav document (nav.xhtml with