v0.2.1 — Metadata patch

Pure metadata patch over v0.2.0. No code changes; runtime behavior is identical.

PyPI refuses re-uploads of an existing version, so this release exists solely to sync the PyPI package page with the GitHub repo's current branding — dropping "Powered by Claude Code" from the tagline and "via Claude Code" from the short description.

Install:

pip install --upgrade animyst

Or fresh:

pip install animyst

Technical references to Claude Code in install requirements, architecture, and safety sections remain — they describe how the tool actually works.

v0.2.0 — Rite framework

ANIMYST pivots from a chat-agent TUI to a CLI for autonomous build rites driven by the Ralph protocol. Describe what you want, walk away, come back to a working repo.

pip install animyst
animyst summon "Landing page for a coffee subscription. Warm orange. Three tiers. Email signup. Mobile-first."

Added

• New animyst CLI surface: summon, status, attach, watch, stop, banish.
• animyst summon "<description>" scaffolds a rite directory, writes a vetted RITE.md (the Ralph-protocol prompt with the user's description interpolated in) and .claude/settings.json (safety deny rules), and starts an autonomous build loop in a detached tmux session.
• animyst watch opens a Textual tracker that polls the rite registry + tmux + git + per-rite state files to render a live cross-rite view.
• loop.sh — bash loop driver that runs claude -p in iteration until the agent prints RITE_COMPLETE or the iteration cap is hit (default 15, configurable via --cap).
• Global rite registry at ~/.animyst/rites.json (override via ANIMYST_DIR).
• Iteration mode: animyst summon "<change>" from inside an existing rite directory treats the description as a change request instead of a fresh build.
• Per-rite WHAT_CHANGED.md updated each loop iteration — plain-English review surface for non-developer users.
• Safety wall: 14+ deny rules in the shipped settings.json blocking git push, git add -A, remote git operations, .env* reads, global installs, and piped-shell installs. Validated by adversarial probe (8/8 blocked).

Changed

• Repositioned the entire project around the user story "describe what you want, walk away, come back to a working repo." Target audience is the lightly technical founder, not the developer using AI to type faster.
• Entry point switched from animyst.app:main to animyst.cli:main.
• README, CLAUDE.md, AGENTS.md all rewritten for the new architecture.

Removed

• The entire v0.1.x chat-agent codebase (~3,500 lines): app.py, llm.py, commands/, domain/, services/, storage/, ui/, cyberpunk.tcss. v0.2 imports nothing from these.
• Chat-agent tests, v0.1 build artifacts, stale HTML preview, stale screenshot.
• Unused dependencies: anthropic, openai, google-genai, gitpython. v0.2 uses claude -p via subprocess (Claude Code subscription) plus the git and tmux CLIs.

Requirements

• Python 3.10+
• Claude Code installed and signed in
• tmux
• git

Validation

Convergence proven by an end-to-end run: plain-English description → 6 iterations → 8 commits → working Next.js + Tailwind + shadcn site → npm run build green. Zero safety violations across all iteration logs. All 8 adversarial probe operations (git push, git add -A, .env reads, global installs, etc.) blocked by deny rules.

Animyst v0.1.1

[0.1.1] - 2026-04-02

This release brings Animyst's public repo, product copy, and recent architecture work into sync.

Added

• Persistent per-agent conversation history with automatic session resume.
• MCP binding with transport-aware persistence and basic health checks.
• Repository-backed storage, service, command, and UI module separation.
• Ralph loop scaffolding for autonomous implementation passes.
• Focused pytest coverage for repositories, chat services, commands, and MCP flows.

Changed

• Reframed the public README and preview docs around what Animyst actually does today.
• Updated package metadata to match the current local-first terminal workspace positioning.
• Refreshed maintainer and Ralph loop documentation for the current roadmap.

Notes

• This is a patch release over v0.1.0 that packages the post-refactor baseline and documentation cleanup.