Skip to content

osModa v1.3.0 — stable core, Claude Code as the main runtime

Latest
Latest

Choose a tag to compare

View all tags
@bolivian-peru bolivian-peru released this 11 Jun 22:27
v1.3.0
11f7c23

[1.3.0] — 2026-06-11 — Stable core

The first stable cut. Core (10 daemons, the modular gateway, agentd contracts + audit
ledger) is stable; Claude Code is the main runtime. Advanced security hardening and the
company-integration layer continue per STATUS.md.

Added

  • Tier-0 PreToolUse approval hook — the default claude-code agent's native Bash/Write/
    Edit now route through agentd's ApprovalGate + hash-chained ledger (closing the gap where
    native tools bypassed the structured-tool blocklist). Ships via install.sh + the NixOS
    module. Implemented + unit-tested; live end-to-end verification on a box pending.
  • Persistent named chats — multiple distinct conversations per server, each with its own
    --resume session, transcript, and native compaction; /chats REST; zero-move migration.
  • Cross-chat + cross-channel awareness — bounded, deterministic digest of other chats'
    notable actions, injected per-turn with a "caught up on N changes" chip.
  • managed-agent runtime driver (prototype) — Anthropic Managed Agents via the ant CLI
    (cloud sandbox; for the untrusted/scale-out tier, not the tier-0 system agent).
  • docs/LIQUID-SOFTWARE.md — the product thesis.
  • CI: full cargo test --workspace (213 tests, was one module) + a no-drift gate
    (scripts/check-drift.sh) that fails if the runtime default or tool count drift across
    install.sh / osmoda.nix / README.

Changed

  • Claude Code is the canonical default runtime across install.sh, nix/modules/osmoda.nix,
    README, and CLAUDE.md (previously the NixOS module defaulted to openclaw — a flake user got a
    different runtime than a curl-installer user). OpenClaw remains a fully-supported peer.
  • README brought current to v1.3 (named chats, managed-agent driver, 92 tools, 247 tests) with an
    honest Safety Model section ("what the default agent bypasses today").
  • Repo root decluttered — internal planning docs moved under docs/planning/.

Fixed

  • Silent total data-loss in the hosted control plane: encrypted stores now use atomic
    tmp+rename writes with a .bak last-known-good fallback (a torn write previously wiped the
    whole order/billing/auth registry).
  • Wedge auto-restart (staleMin ReferenceError) — fleet self-healing now actually fires.
  • npm test (gateway) runs under Node 24 as well as the pinned 22.

Verification for this release: nix flake check passes on a clean checkout for the first time (4 pre-existing eval breakages fixed; verified on a real box against this exact tree) · 213 Rust workspace tests + 40 gateway tests green in CI · a no-drift gate now fails CI if the runtime default or tool counts ever diverge again.

Full changelog: CHANGELOG.md

Assets 2
Loading