Releases: gogooma125732/parsewall
Release list
Parsewall v1.0.0
Parsewall v1.0.0
Unified release of Parsewall across GitHub Marketplace, PyPI, Codex Marketplace/MCP, Docker Compose, and offline Docker deployment.
GitHub Action
- uses: actions/checkout@v4
- uses: gogooma125732/parsewall@v1
with:
input: path/to/document.pdf
fail-on: reviewThe Action exposes risk-level and the fixed Parsewall JSON result as outputs.
Python CLI and MCP
Published on PyPI: https://pypi.org/project/parsewall/1.0.0/
python -m pip install parsewall==1.0.0
parsewall scan --input ./report.pdf
parsewall-mcpCodex plugin marketplace
uv tool install parsewall==1.0.0
codex plugin marketplace add gogooma125732/parsewall --ref main
codex plugin add document-injection-firewall@parsewallStandalone Codex plugin and marketplace ZIP files are attached below.
Docker and offline deployment
The attached Compose bundle installs from source. The offline Docker bundle includes the Linux ARM64 image and deployment files.
See deploy/LOCAL_INSTALL.md and deploy/OFFLINE_INSTALL.md for installation and verification instructions.
Security model
Parsewall deterministically scans untrusted documents before model access. Unsupported, corrupt, encrypted, incomplete, or dependency-blocked inputs fail closed. A low result does not make document content trusted.
Verification
- 182 automated tests passed
- Python wheel and source distribution published to PyPI
- Release checksums verified
- Action tags
v1.0.0andv1point to the unified release commit
Parsewall v0.1.0
Parsewall v0.1.0
Deterministic, fail-closed inspection for untrusted files before they reach an LLM.
Highlights
- Isolated parser-first inspection
- Fixed JSON result contract
- Network-isolated Docker worker
- Low-only marked text derivative
- Text, Markdown, HTML, DOCX, PPTX, XLSX, PDF, PNG and JPEG support
- Codex Skill, PreToolUse Hook and optional MCP server
Distribution packages
- Docker Compose local installer
- Offline Docker bundle for Linux ARM64
- Python wheel and source distribution
- Codex Plugin and local Marketplace bundles
Security model
A low result does not make document content trusted. Parsed content must continue to be handled as untrusted data, never as instructions.
Unsupported, corrupt, encrypted, incomplete or dependency-blocked inputs fail closed.
Verification
- 180 tests passed
- Ruff and mypy passed
- Plugin and Skill validators passed
- SHA-256 release checksums verified
Platform note
The included offline Docker image targets Linux ARM64.