chore(release): version packages

[openspec-release-bot]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@fission-ai/openspec@1.8.0

Minor Changes

• #1303 1aa0f2a Thanks @solanab! - Add the vendor-neutral agents target: openspec init --tools agents installs the workflow skills to .agents/skills/openspec-*/SKILL.md, the shared location AGENTS.md-compatible assistants read. It is skills-only, so no slash commands are generated. Because agents is now a real target, --tools all includes it and creates .agents/skills/ where it previously did not.

• #1274 7a4a745 Thanks @NicoAvanzDev! - Generate GitHub Copilot coding agent setup and custom agent files during openspec init and keep them synchronized during openspec update.

• #1214 161f945 Thanks @showms! - Add MiniMax Code as a global skills-only tool target.

• #1518 568e56c Thanks @clay-good! - ### New Features

  • Atlassian Rovo Dev CLI — openspec init --tools rovodev installs the OpenSpec workflow skills for Atlassian's Rovo Dev CLI. It is skills-only (no slash commands), written to .rovodev.

  Bug Fixes

  • Codex skills now live in the shared .agents directory — openspec init and openspec update install Codex skills under .agents/skills/ (the canonical location assistants read) and migrate an existing .codex skills directory in place. Files you customized are preserved, not overwritten.
  • openspec status separates planning from implementation — status now reports isPlanningComplete (every non-skipped planning artifact exists; skipped artifacts count as satisfied without being written) distinctly from overall progress, and its messages no longer imply a change is finished before it has been implemented. isComplete is kept as a compatibility alias, so existing scripts keep working.

• #1517 73207a6 Thanks @clay-good! - Make GitHub Copilot cloud coding-agent files opt-in. Selecting the github-copilot tool no longer silently writes a GitHub Actions workflow into .github/; openspec init now asks first (default No) and remembers the choice in openspec/config.yaml (githubCopilot.cloudAgent). Use --copilot-cloud / --no-copilot-cloud to decide non-interactively.

  • openspec update never prompts — it only refreshes cloud files for projects that opted in (or that already have generated cloud files, so existing setups keep working).
  • Opting out (--no-copilot-cloud or cloudAgent: false) removes OpenSpec-managed cloud files; a user-customized file is always preserved, never overwritten or deleted.
  • init and update now report whether cloud files were written, skipped, or left untouched — and if you already have your own copilot-setup-steps.yml, they say it was preserved and that you need to add the OpenSpec install step by hand.

• #1484 521ee33 Thanks @clay-good! - Retire a capability when a change removes its last requirement. A change that declares retire_capabilities: true in its .openspec.yaml (alongside the schema: that file requires) may now be archived even when its REMOVED entries take a capability's last requirement: openspec archive deletes that capability's main spec instead of aborting with "Spec must have at least one requirement". Without the marker nothing changes — the archive aborts exactly as before, except the message now names the marker as the way out. Retirement happens only when the emptied spec could not have been written at all, every one is named in the archive output, a pasteable git checkout is included when the spec lived in the caller's checkout, and --no-validate never retires. Archive now also rejects a main spec with duplicate canonical requirement names instead of letting delta reconciliation collapse one of the duplicate blocks. One thing to know before retiring: a capability's spec is the base another change's MODIFIED block is checked against, so an in-flight change that modifies the capability you just retired will keep validating clean and then refuse to archive ("target spec does not exist; only ADDED requirements are allowed for new specs") — close or rework that change alongside the retirement.

Patch Changes

• #1502 ece8660 Thanks @clay-good! - openspec validate now treats the English SHALL/MUST convention as guidance in normal mode, so requirements written in other languages can validate. Strict mode continues to enforce the convention.

• #1483 2b3d368 Thanks @clay-good! - Tell the caller which flag to pass when openspec archive cannot ask its confirmation questions. An AI agent (or any script) runs the CLI with stdin closed, so every prompt rejects with @inquirer's User force closed the prompt with 0 null — the archive aborted with an error that named neither the question nor the flag, and agents burned a turn guessing (#1479). Each confirmation now reports what it needed and a pasteable rerun that carries the flags you already passed: openspec archive <name> --skip-specs --yes stays a --skip-specs run, so following the suggestion cannot merge specs you opted out of merging, and a change name that needs quoting gets double quotes, the one form bash, zsh, PowerShell and cmd.exe all read the same way (a name no shell reads literally even quoted — one containing $, a backtick, or the %/! that cmd.exe still expands inside quotes — is left as a <change-name> placeholder rather than a command that would target something else). openspec archive with no change name used to swallow the same failure, print No change selected. Aborting. and exit 0 — success for a run that archived nothing; it now exits 1 asking for a change name, matching how openspec show and openspec validate already behave without a terminal. The check is reactive — it inspects a prompt that already failed — so answers piped into the command, --yes, --json, and Ctrl-C all behave exactly as before, and a run that OpenSpec already considers non-interactive (CI, OPEN_SPEC_INTERACTIVE=0, --no-interactive) gets the guidance even when the runner allocated a pty. The onboarding walkthrough, the only generated guidance that tells an agent to run openspec archive, now shows --yes.

• #1486 427abf4 Thanks @clay-good! - Task progress now counts indented sub-tasks. A tasks.md whose sub-tasks were unfinished reported ✓ Complete in openspec list and openspec view, was missing those tasks from the openspec instructions apply list, and archived with no incomplete-task warning, because both checkbox parsers only matched checkboxes at column 0.

  Progress counting and the apply task list now share one parser, so list, view, archive and apply agree about which lines of a tasks file are tasks. A checkbox with no text after it is left out of the apply list, which has nothing to act on, but still counts toward every progress number; a file of nothing but such checkboxes now asks to be rewritten rather than reporting itself done. The shared pattern matches every line the two it replaced matched, and more, so task counts can rise but never fall: no change starts reporting less work than before, and archive's incomplete-task warning can only become stricter. Checkboxes are still counted wherever they appear, including inside a code fence, an HTML comment or an indented block, so a tasks.md that shows a checklist as a format example can now count that example as work — remove it from the file, or pass --yes to archive.

• #1500 26bd1d4 Thanks @clay-good! - Keep generated workflows on the selected store, handle optional workflow fallbacks safely, and validate synced specs before reporting success.

• #1490 45cca5d Thanks @clay-good! - Say before confirmation when archiving a change will delete a note written next to a requirement. A requirement absorbs anything below it that OpenSpec doesn't recognize as a new heading — a note indented by the one to three spaces Markdown allows, for example — so removing or modifying that requirement took the note with it, silently. openspec archive now names content the rebuilt spec would actually drop and where to move it to keep it. The merge itself is unchanged: nothing is relocated, because a # line inside a scenario looks identical to a note and moving one of those would rewrite the spec wrongly.

• #1492 690a27e Thanks @mc856! - openspec init and openspec update no longer delete the CoStrict and Junie command files they just generated. Legacy cleanup removes artifacts older OpenSpec versions left behind, and two of its patterns named paths the current adapters still write to. CoStrict's was a whole-directory removal of .cospec/openspec/commands/, the folder the adapter writes opsx-<id>.md into, so every run wiped the directory — including any file the user kept there — while the banner above it read No user content to preserve. Junie's .junie/commands/opsx-*.md listed its own current output. Cleanup runs before the config migration, so on a config that has no profile key yet the missing command files make delivery detection read the project as skills-only and persist that to the global config: the files are not regenerated, and the preference changes for every other project too.

  CoStrict is now a file pattern, .cospec/openspec/commands/openspec-*.md, matching the three commands the pre-opsx CoStrict integration wrote there (openspec-proposal.md, openspec-apply.md, openspec-archive.md) and the same shape every other file-based tool already uses. Junie's entry is removed outright: Junie support arrived after the slash configurators that wrote openspec-* files were deleted, so no OpenSpec version ever created those files there. Genuinely legacy files are still detected and removed, and no other tool's patterns change — they never overlapped their adapter's current output.

• #1501 0b20ae3 Thanks @clay-good! - Keep the propose workflow focused on planning, clarify material ambiguities before creating a change, and hand implementation off to the apply workflow.

• #1503 8a3850d Thanks @clay-good! - When exploration turns into a new change, generated explore guidance now instructs agents to run openspec new change before writing requested artifacts. This preserves the required .openspec.yaml metadata instead of letting an agent create an incomplete change directory by hand. After the user accepts a capture, explore also creates the requested artifacts without requiring another workflow command.

• #1513 622c509 Thanks @FasterPHP! - Honor telemetry.enabled in global config. false disables anonymous telemetry and openspec update version checks; unset keeps telemetry enabled, and env/CI opt-outs still take precedence.

• #1499 9cd845f Thanks @clay-good! - Keep generated files, specs, archive moves, and local state inside their intended security boundaries without breaking linked monorepo workflows.

• #1482 84ebc57 Thanks @clay-good! - openspec validate <change> now reports a MODIFIED requirement that omits a scenario the main spec still has — the same loss archive already refuses to apply — so the change fails at authoring time instead of at archive time. A change carrying a stale MODIFIED block will start failing validation; it was already unarchivable, and the message names the scenarios to copy back in.