Reference · Changelog

Changelog

A hand-authored record of what changed, newest first. It tracks releases, not commits, so it leans toward what changed for you. The repo's CHANGELOG.md has every entry.

0.4.2 — 2026-11-06

A global OpenCode install. npm install -g agentry-opencode now puts agentry on PATH (beside agentry-opencode), and agentry init --host opencode --global --apply writes the agents, commands and skills into OpenCode's global config dir instead of a repo's .opencode/.

0.4.1 — 2026-11-06

agentry init --host opencode --apply in a repo with no process file yet no longer exits with an error. It keeps the payload and the vendored CLI, skips the AGENTS.md block with a warning that names /agentry/init, and exits 0. Run /agentry/init in OpenCode, then run the init again.

0.4.0 — 2026-11-06

Agentry 0.4 runs on OpenCode as well as Claude Code. The agentry-opencode npm package gives OpenCode native board, lane and memory tools, and agentry init --host opencode writes a repo's agents, commands and skills, generated from the same plugin/ source as the Claude Code plugin. The memory index no longer needs node:sqlite, so agentry mem also runs under Bun. The lane exec reports, restarts, resumes and briefs say more about what a round did and what comes next.

To upgrade on Claude Code, remove the marketplace, add it again with #v0.4.0 and restart. On OpenCode, run opencode plugin add agentry-opencode, opencode service restart and npx agentry-opencode init --host opencode --vendor in the repo (add --apply to write; --vendor copies the CLI into .agentry/, so the agents' CLI fallback works without a checkout). The repo's CHANGELOG.md lists what to change when upgrading from 0.3.

0.3.0 — 2026-11-05

Agentry 0.3 turns the plugin into a harness for a repo's own board. /agentry:exec builds a task whose contract is already written, in its own git worktree. A second agent checks the work against the acceptance, and /agentry:merge lands it only past your repo's gate. Each step is recorded in a ledger, so a cold session can resume. /agentry:init learns the board you have or scaffolds one, and /agentry:start and /agentry:close open and end a session the same way each time. The agentry command does the mechanical work; the plugin starts no server and ships no hooks. Memory is now agentry mem, over the same files as before, and the plugin's own servers are gone.

Restart Claude Code after updating. agentry --version prints the build you are running. The repo's CHANGELOG.md lists what to change when upgrading from 0.2.

Added

  • Your board, as lanes — /agentry:exec builds a contracted task in its own worktree with a builder, an independent verifier and your repo's gate; /agentry:merge lands it.
  • Setup and sessions — /agentry:init learns your board or scaffolds one; /agentry:new, /agentry:board, /agentry:start, /agentry:close and /agentry:ask.
  • The agentry CLI — board, lane, mem, draft, run and lock, on the Bash tool's PATH while the plugin is enabled.
  • Work already on main — agentry lane merge <id> --resolve landed closes a verified lane whose commits reached main another way, found by patch.
  • Time limits and locks — agentry run --limit <dur> stops a command's whole process group at the limit; named locks share a slot across worktrees.
  • Workflow presets — exec.preset (lean, standard, strict) sets how much checking a lane gets.

Changed

  • Memory is the agentry mem CLI — the memory MCP server is gone and the plugin starts no server. Existing stores read unchanged. A lane's close writes its episode in one call.

Removed

  • All plugin hooks, the primer among them. Their work moved into the commands.
  • The run-state server from v0.2. /agentry:go and its steps write the spec, plan and task files as plain files under .agentry/work/.

v0.2

The release that adds a multi-page docs site and a self-eval harness to the conductor-plus-memory core.

Added

  • The docs & marketing site — this multi-page docs area, covering Getting Started, Capabilities, and Reference.
  • The self-eval harness — decision-quality measures, controls, and a corrections log.

v0.1 → v0.1.2 — initial Agentry

The first releases established the core: a right-sized conductor, the SDLC specialist roster, durable memory, and the primer that warms a session.

Added

  • The right-sized conductor — routes a task to the least process that wins (one-shot → spec-first → decompose + verify).
  • The SDLC specialist roster — the explorer, architect, implementer, verifier, librarian, and the rest of the specialist agents.
  • The memory MCP — durable facts and episodes the conductor can recall, so run #2 is warmer than run #1. Replaced by agentry mem in 0.3.
  • The primer — a minimal session nudge that points the conductor at the repo's conventions and memory.