mylesmcook/thermos
v0.3.1MIT
Independent paired or adversarial reviews with shared scope and evidence verification.
Changelog
0.3.1 - 2026-11-08
- Add an explicit Claude manifest for claude.ai marketplace synchronization.
0.3.0 — 2026-09-09
- Consolidate three-lens adversarial review into Thermos; share scope and evidence verification. Preserve paired and single-lens behavior.
0.2.1 - 2026-08-24
- Added the portable Agent Plugins 1.0.0 manifest while retaining the Codex compatibility overlay.
0.2.0 - 2026-07-10
- Made Thermos use two ordinary same-wave native subagents with explicit correctness/security and maintainability contracts.
- Added context-free child starts where the host exposes
fork_turns. - Removed the sequential surrogate so Thermos never calls a non-parallel review a paired review.
- Clarified inherited permission behavior and added native-child smoke assertions.
0.1.0 - 2026-07-09
- Created the Codex-native, skills-only Thermos plugin.
- Added two parallel read-only review passes with one verified synthesis.
- Defined branch, pull-request, explicit-range, and dirty-worktree scope rules.
- Added an explicit parent-owned repair loop for
work until Thermos is happy. - Added skill metadata, OpenAI-backed design notes, and maintainer smoke cases.
- Pinned upstream attribution and preserved both MIT copyright notices.