Skip to content

Releases: JinNing6/CyberHuaTuo-Plugin

CyberHuaTuo v0.2.5

Choose a tag to compare

@JinNing6 JinNing6 released this 16 Jul 06:15

MCP Registry-ready release

  • Adds the stable cyberhuatuo mcp stdio server entry point.
  • Publishes official MCP Registry metadata as io.github.JinNing6/cyberhuatuo.
  • Keeps the Gold-first diagnosis, Agent Action Guard, contribution, and Soul Ring flows available through one package.
  • Aligns Claude, Codex, MCP, PyPI, and public IssueOps version surfaces.

Quick start

pip install -U cyberhuatuo==0.2.5
cyberhuatuo guard --self-test --workspace-root .
cyberhuatuo mcp

CyberHuaTuo v0.2.4: Trusted Cure Integrity

Choose a tag to compare

@JinNing6 JinNing6 released this 15 Jul 08:45

Trusted cure integrity

CyberHuaTuo 0.2.4 makes the no-Key cure path evidence-aware and explicit about trust.

  • Added cyberhuatuo cure: Gold first, then at most one clearly labeled Reviewed candidate; Draft is never presented as a cure.
  • Replaced whole-document token coverage with stable error signatures and distinctive-entity matching.
  • Added local-only cure-feedback <case-id> yes|partial|no; it stores no original traceback and never auto-promotes a case.
  • Added path-confined, exclusive-create prescription storage to prevent traversal and silent overwrite.
  • Hardened Reviewed/Gold evidence, reviewer, verification, safety, rollback, and date gates.
  • Added 8 Reviewed prescriptions from previously solved maintainer incidents and one reproducibly verified Gold LangChain prescription.
  • Fixed Soul Ring promotion metadata and kept Draft contributions outside trusted retrieval and rewards.

Current bundled inventory: 53 cases: 1 Gold, 8 Reviewed, 44 Draft.

Install and verify:

python -m pip install -U cyberhuatuo==0.2.4
cyberhuatuo cure "ImportError: cannot import name 'ChatOpenAI' from 'langchain.chat_models'"
cyberhuatuo quality-audit

A Reviewed result is a candidate to verify, not an instruction to execute. CyberHuaTuo does not run the returned fix.

CyberHuaTuo v0.2.3: Privacy-first Guard Case Intake

Choose a tag to compare

@JinNing6 JinNing6 released this 14 Jul 09:36

Privacy-first Guard case intake

CyberHuaTuo 0.2.3 turns a real Guard decision into a reviewable local case without executing the proposed command or uploading the report.

  • Added --expected ALLOW|ASK|BLOCK for an explicit expected-vs-actual contract.
  • Added preview-first --report, repeated --redact, explicit confirmation, no implicit overwrite, and fail-closed handling for private-key material.
  • Added public forms for false positives, ordinary false negatives, and integration gaps.
  • Added private vulnerability reporting for reliable Hook, parser, wrapper, protocol, or enforcement bypasses.
  • Kept the existing no-expectation exit contract unchanged: ALLOW=0, ASK=2, BLOCK=3.

Install and verify:

python -m pip install -U cyberhuatuo==0.2.3
cyberhuatuo guard --self-test --workspace-root .
cyberhuatuo guard "git status --short" --workspace-root . --expected ALLOW --report guard-case.md

The Guard is a deterministic guardrail, not a sandbox or complete reference monitor. No external adoption or accepted external case is claimed by this release.

CyberHuaTuo v0.2.2: Agent Action Guard

Choose a tag to compare

@JinNing6 JinNing6 released this 13 Jul 07:43

What is new

  • Agent Action Guard for destructive shell actions through CLI, MCP, Codex/Claude plugin skill, and PreToolUse hook.
  • Zero-execution onboarding check: cyberhuatuo guard --self-test --workspace-root ..
  • Stable ALLOW / ASK / BLOCK decisions with rule IDs, resolved targets, reversibility, and safer next actions.
  • Four Windows network diagnosis and optimization prescriptions with explicit diagnosis, verification, and rollback boundaries.
  • Version-consistent PyPI, Claude, Codex, README, and IssueOps entrypoints.

Safety boundary

The guard is a high-value guardrail, not an operating-system sandbox or complete reference monitor. It never executes or rewrites the reviewed command. Host paths that do not emit PreToolUse remain outside hook coverage.

Verify

python -m pip install --upgrade "cyberhuatuo==0.2.2"
cyberhuatuo guard --self-test --workspace-root .

22-second zero-execution demo

Watch the Agent Action Guard demo or open the full-resolution cover.

The video runs the real cyberhuatuo guard --self-test --workspace-root . flow and shows ALLOW, ASK, and BLOCK. The displayed command strings are reviewed only; none of them is executed.

CyberHuaTuo v0.2.1

Choose a tag to compare

@JinNing6 JinNing6 released this 07 Jun 06:42

CyberHuaTuo v0.2.1 patch release.

Highlights:

  • Adds chat-visible Soul Ring PNG/GIF visual artifacts for Codex and Claude conversations.
  • Fixes GIF LZW encoding so generated animations render correctly in chat clients.
  • Updates Codex/Claude skills, MCP routes, CLI/docs, and release contracts for the v0.2.1 install loop.

Local gates passed before publication:

  • ruff check .
  • pytest: 212 passed
  • CLI and MCP import smoke checks: pass
  • build sdist/wheel: pass
  • release boundary: pass

CyberHuaTuo v0.2.0

Choose a tag to compare

@JinNing6 JinNing6 released this 05 Jun 07:35

CyberHuaTuo v0.2.0 public growth loop release.

Local gates passed before publication:

  • ruff check .
  • pytest: 205 passed
  • launch-assets: pass
  • build sdist/wheel: pass
  • release boundary: pass
  • twine check: pass
  • claude plugin validate: pass
  • mcpb validate/pack: pass