Free2PA v0.4.2

Free2PA v0.4.2\n\n- Hello World demo starts at the root page.\n- Unsigned saved revisions stay pending while the app continues with the last signed control file.\n- Tooltip popups now have close controls and consistent check typography.\n- Scan and GitHub Action cover agent control files such as SOUL.md, AGENTS.md, TOOLS.md, policies, prompts, and any file with a .c2pa.json sidecar.\n- Package excludes Workbench, private keys, env files, IDE metadata, and unrelated radio intern files.\n\nTarball SHA1: bf0463e91f2ecf264e6688741ad82238d6f35b1a\nTarball SHA256: 434df1cad6adc8e95522e063c98905675443c9bc48fc50a6d37e91c36c39a97a

Free2PA v0.4.1 - Toolkit-first Build Week release

Free2PA is an Apache-2.0 developer toolkit for verifier-local publisher trust and tamper-evident AI agent Nerve Center files. This revision restores the generic Agentic Factory as the primary judge experience and moves Hello World to a separate reference integration. It includes the CLI, Node load gate, HTTP API, MCP verify_asset tool, GitHub Action, guarded repair, installable Codex skill, 21 passing tests, refreshed judge documentation, a 2:55 toolkit-first video, and a 15-image collateral set.\n\nFree2PA uses C2PA-style provenance concepts for origin and edits, but is not a conforming or interoperable C2PA implementation.

Free2PA v0.4.0 - Protected Hello World Agent

Free2PA v0.4.0 adds a complete, runnable Azure-hosted Hello World agent that makes the load boundary visible. Its signed SOUL.md allows only optimistic greetings. The comparison lane shows a bitter edit reaching an unchecked model; Block prevents protected inference, and Repair verifies and recovers the signed optimistic soul before a fresh model call. The release also includes the real HTTP comparison endpoint, three signed trust fixtures, 20 automated tests, a first-run example, and an explicit favicon for a clean judge experience. Core signing, verification, trust, MCP, load gate, and repair remain model-independent. Operators provide their own optional OpenAI or Azure OpenAI account.\n\nSHA-256: 55d141b24ae36900c559c01445c820f04a6bf05bde92501e8c3e5c8b017b48bb

Free2PA 0.3.3 - Optional installable LLM auditors

Free2PA 0.3.3 separates deterministic trust from optional LLM behavioral review.\n\nHighlights:\n- Free2PA signing, verification, publisher trust, repair, CLI, MCP, and load gates run with no LLM account.\n- Operators can bring their own OpenAI, Azure OpenAI, or OpenAI-compatible account.\n- Other hosted or local LLMs can be installed through the FREE2PA_AUDITOR_MODULE provider contract.\n- New free2pa auditor status command makes the optional configuration visible.\n- Provider-neutral workbench labels retain provider and model evidence in results.\n- An LLM report cannot change REJECT to LOAD, admit a publisher, sign, or repair.\n- 18 automated tests pass and the clean dependency audit reports zero vulnerabilities.\n\nPackage SHA-256: 07f7a3e00c18cb342ea0bfd168af40fd8c436a84a4e2fdf1849d997d3e3e36c2\n\nFree2PA is C2PA-inspired and uses its own sidecar format; it does not claim C2PA conformance.

Free2PA 0.3.2 - Download-to-done load gate

Free2PA 0.3.2 makes the developer path concrete from repository discovery to runtime enforcement.\n\nHighlights:\n- New framework-neutral free2pa/load-gate API returns verified content or throws before an agent can consume it.\n- Codex skill now fact-gathers the target harness and reports its entry point, Nerve Center, load boundary, security checks, integration surface, and unknown owner decisions before editing.\n- README now begins with runnable Codex and manual setup paths.\n- Complete implementation runbook covers custom Node harnesses, OpenClaw or other fixed loaders, MCP-capable frameworks, tests, and CI.\n- free2pa --help now works as a first-run discovery command.\n- 16 automated tests pass; clean dependency audit reports zero vulnerabilities.\n\nPackage SHA-256: d35030e15d56e36052d94c0a5939f036cfd59dac33d4c4c34eca983bca872f44\n\nFree2PA is C2PA-inspired and uses its own sidecar format; it does not claim C2PA conformance.

Free2PA 0.3.1 - Agent-native MCP trust gate

Free2PA 0.3.1: Agent-native MCP trust gate

Free2PA is the automatic trust gate for the files that steer AI agents. This
patch makes the Streamable HTTP MCP server a first-class integration surface.

Added

• Generic verify_asset MCP tool for any Nerve Center file and neighboring
  Free2PA sidecar.
• Declared MCP input and output schemas.
• Structured PASS/FAIL and LOAD/REJECT results with independent
  signature, exact-file, certificate-currentness, and local publisher-trust
  facts.
• Stable VERIFIED, CONTENT_CHANGED, INVALID_SIGNATURE, EXPIRED_CERT,
  and UNTRUSTED_ISSUER reason codes.
• Read-only, idempotent MCP tool annotations and bounded request bodies.
• A complete public project dossier with architecture, threat model, judging
  strategy, evidence, schedule, risks, and definition of done.

Verification

• 14 automated tests pass.
• A real Streamable HTTP client proves trusted LOAD, changed-file REJECT,
  and outside-group REJECT behavior.
• The exact public release commit passed GitHub-hosted CI.
• The archive clean-installs on Node.js 20+ and includes the Codex skill.
• npm audit --omit=dev reports zero known vulnerabilities.
• No signing private keys are included.

Free2PA is Apache-2.0 freeware. It is C2PA-inspired, not a conforming or
interoperable C2PA implementation, and does not claim C2PA endorsement.

Free2PA 0.3.0 - Agentic Factory

Free2PA 0.3.0: Agentic Factory

Free2PA is Apache-2.0 freeware that makes the files steering an AI agent
tamper-evident for a project-defined, ad-hoc trust group.

New in this release

• Live Agentic Factory: Run the same trusted, changed, or outside-group
  control file through agent lanes with and without Free2PA.
• Programmatic response policies: Agent hosts can block, alert and
  continue, log, or restore a known trusted signed version.
• Guarded repair: free2pa repair restores only from a valid, current,
  locally trusted signed receipt and preserves the rejected file as evidence.
• Installable Codex skill: free2pa codex-skill install adds the included
  free2pa-protect-agent skill so Codex can retrofit the trust gate into an
  existing agentic application.
• Preserved research workbench: The original sign, verify, and GPT-5.6
  audit interface remains available at /workbench.html.

Install

npm install ./free2pa-0.3.0.tgz
npx free2pa codex-skill install

Public CI for the release source passed all 13 tests:
https://github.com/kilroyblockchain/free2pa-devtool/actions/runs/29644256018

Free2PA is C2PA-inspired but is not a conforming C2PA implementation and does
not claim C2PA interoperability or endorsement.

Free2PA v0.2.1 - OpenAI Build Week

Final Developer Tools submission release. The Azure page is a reference verifier and judge sandbox; the toolkit in this repository is the product developers install and integrate.\n\n- Installable CLI for publisher identity, signing, trust admission, verification, scanning, auditing, and serving\n- Browser, HTTP, MCP, and reusable CI action surfaces\n- GPT-5.6 behavioral auditing with Azure managed identity support\n- 13 automated tests and a verified 35 KB package\n- Read-only public judge demo and explicit Build Week provenance record\n\nThe attached tarball was installed in an empty directory and passed key generation, signing, trust admission, and verification. Public Ubuntu CI passed on the exact tagged commit: https://github.com/kilroyblockchain/free2pa-devtool/actions/runs/29641643306\n\nJudge guide: https://github.com/kilroyblockchain/free2pa-devtool/blob/main/docs/JUDGE_GUIDE.md\n\nBackup demo video: https://github.com/kilroyblockchain/free2pa-devtool/releases/download/v0.2.1/Free2PA-Build-Week-demo.mp4

Free2PA v0.2.0 - Build Week

First freeware developer-tool release. Free2PA forms ad-hoc publisher trust groups for AI agent skills and enforces signature, content-integrity, certificate-validity, and local-trust gates through a CLI, GitHub Action, browser, HTTP API, and MCP. GPT-5.6 adds an independent structured behavioral security audit. The public Azure demo uses scoped managed identity and stores no model API key.