Skip to content

Releases: charles-chenzz/cathy-release

Cathy v0.1.0-alpha.1

Cathy v0.1.0-alpha.1 Pre-release
Pre-release

Choose a tag to compare

@charles-chenzz charles-chenzz released this 03 Sep 01:45
a005617

[0.1.0-alpha.1] - 2026-09-03

Added

  • Interactive terminal agent with streaming responses, native file and Bash
    tools, persistent threads, model selection, and thread continuation.
  • Anthropic, OpenAI, OpenAI-compatible, ChatGPT/Codex, and GitHub Copilot
    provider integrations using API-key or OAuth authentication as applicable.
  • Anthropic Messages, OpenAI Chat Completions, OpenAI Responses, and Codex
    Responses protocol support.
  • Optional MCP tool discovery and execution over stdio and Streamable HTTP.
  • Lazy-loaded project and user skills from .agents/skills and
    .claude/skills.
  • Metadata-only local JSONL telemetry and opt-in OTLP export.
  • Linux, macOS, and Windows release archives for amd64 and arm64, with SHA-256
    checksums and per-archive SBOMs.

Changed

  • The module now uses github.com/charles-chenzz/cathy, and the user-facing
    command is built from cmd/cathy as cathy.
  • Release builds expose their tag through cathy --version.

Security

  • Updated the Go toolchain and reachable dependencies to remove known reachable
    vulnerabilities detected before this release.
  • Added vulnerability scanning to CI and the release quality gate.
  • Cathy-created credential and state files use mode 0600; native file tools
    enforce workspace boundaries, and every Bash command requires approval.

Known limitations

  • Alpha storage and configuration formats may change incompatibly.
  • Approved Bash commands execute locally without an isolation sandbox.
  • Credentials are stored in a local permission-restricted file rather than an
    operating-system keychain.
  • macOS and Windows binaries are not yet code-signed or notarized; Windows is
    experimental pending broader real-device testing.
  • There is no automatic updater.
  • Cathy's ACP and MCP surfaces intentionally implement only the subsets listed
    in the project README.