v0.1.3: clean failed renders and refresh compatibility

Removes partial PNG files after failed or cancelled PDF rendering. Refreshes plugin discovery metadata and adds verified DSH 0.2.1-alpha.2 compatibility.

Validation: 30 packed-plugin tests passed, including a reproduced failed-render regression. Post-merge CI passed: Linux/macOS package tests, three DSH release versions, and disposable DSH 0.2.0-rc.2 CLI install, startup, tool execution and uninstall. Every archive file was verified byte-for-byte against the merged commit. Archive, SHA-256 checksum and release manifest are attached.

Requires external Poppler; OCR additionally needs Tesseract and language data. The plugin reads files and runs native utilities with a restricted inherited environment. Windows, rollback and packaged desktop GUI checks remain unverified. Store admission is still subject to review of these documented capabilities.

v0.1.2: store compatibility and bounded native environment

DSH STORE #1336 reported a missing compatibility declaration and file/command/environment capability signals.

This release adds the store-specific compatibility range, exact DSH release matrix and separately measured CLI install/start/uninstall evidence. Native Poppler/Tesseract subprocesses now receive only required runtime-path, font and OCR environment settings instead of inheriting host API keys and other credentials. File and command access remain explicitly documented.

Validation:

• 29 packed-plugin tests, including OCR, rendering, filesystem denial, cancellation and environment filtering.
• Actual disposable DSH 0.2.0-rc.2 profile: official CLI install, startup, five tools, PDF extraction, uninstall and clean restart.
• DSH STORE submission preflight passed for this fixed commit: complete 5/5-file scan, zero critical findings, one child-process review warning. Reports are attached.
• CI results.

This is source remediation and author verification, not store approval or an independent security audit. The store's stricter automatic admission policy still flags required file/command/environment access. Review evidence and limitations: Store #1336 record.

The release also includes the previously merged bounded PDF search and inspection-cache reuse improvements. Existing custom native wrappers that depend on arbitrary environment variables must set those values in the trusted wrapper itself.

Install the archive in the profile your DSH instance actually runs. Poppler and optional Tesseract language data remain external dependencies. No npm publication or change to the user's live profile is performed by this release.

v0.1.1: automatic OCR for mixed PDF pages

Automatic OCR now checks for embedded images as well as sparse text. A page containing selectable prose above a scanned receipt no longer silently skips the receipt.

• Preserve the original selectable text and append a clearly labeled OCR reading on mixed pages. Duplicate readings are identified explicitly.
• Report unavailable image detection and OCR. Empty OCR output cannot erase selectable text.
• Expose embedded-image counts during inspection and warn about equation and table structure that plain text cannot preserve.
• Keep processing local, serialized, cancellable and bounded. No cloud parser, background model download or desktop modification is required.

Requires Poppler's pdfimages in addition to the existing native tools. Logos and searchable scans can trigger extra OCR; ocr: "off" bypasses it.

Validation: all 22 packaged integration tests passed on Linux (Node 22 and 24) and macOS (Node 24). The synthetic eight-page comparison recovered all three previously missed mixed-page receipt facts without changing other extraction scores. General mathematical notation recognition remains unverified and requires visual checking or a specialized parser. Packaged desktop GUIs and Windows plugin execution were not tested.

Install the attached archive using the plugin manager for your existing Harness profile. This release does not change a running installation automatically.

DSH PDF Reader v0.1.0

Read searchable and scanned PDFs in DeepSeek Harness through five tools: inspect, read, search, render and dependency status. Page images and crops are returned as Harness image attachments when the routed model supports vision. English and Simplified Chinese OCR run locally with Tesseract.

Install the attached .tgz through the desktop Plugins page and enable dsh-pdf-reader. Install Poppler and Tesseract separately; see the README for platform setup and official desktop profile instructions. No npm runtime dependencies or model weights are included.

Validation covers real Harness services, OCR, citations, truncation, rendering/crops, permissions, cancellation and explicit failure reporting. Official desktop compatibility is based on shared APIs and its documented plugin mechanism; its packaged GUI and Windows are not yet verified. OCR and complex visual interpretation remain fallible, and no model-accuracy improvement percentage is claimed.

All 15 integration tests passed in each of three CI jobs: Linux with Node 22 and 24, and macOS with Node 24. See CI evidence.