- Drop the @opencode/plugin dependency: Plugin.define is an identity
function, and both hosts load a plain { id, setup, server } object.
This removes ~3,700 lockfile lines from every OpenCode 1 and 2 install.
- Remove the OpencodePowerPack named export. OpenCode 1.18.7+ uses the
default export's server() and never scans named exports when it exists.
- Derive the OpenCode 2 rulesets from the OpenCode 1 permission maps so
the read-only policy has a single source, and parse SKILL.md once.
- Build fresh agent definitions per hook call. OpenCode 2 appends global
rules to agent permissions in place, and the shared module array leaked
those rules between code-explorer and code-architect.
- Un-export eval runner helpers that nothing imports.
Record the 1.18.32 and 2.0.14 host inspections, describe the shared
plugin hook in the notices, and sync the lockfile after restoring the
1.18.7 engine floor.
OpenCode 1 calls the default export's server() hook while OpenCode 2
calls setup(), so a single module can support both:
- Keep the original config hook (config.skills.paths and config.agent)
and expose it through server() plus the legacy OpencodePowerPack named
export for releases that only discover named exports.
- Keep the V2 skill and agent transforms for OpenCode 2 unchanged.
- Build both the V1 permission maps and the V2 ordered rulesets from the
same SKILL.md frontmatter.
- Extend the smoke test to verify both hosts: the OpenCode 1 path asserts
skill-derived commands and agent permissions from the REST API, and the
OpenCode 2 path keeps the verifier plugin.
- Run the smoke matrix on 1.18.7 and 2.0.14, and advertise >=1.18.7.
The OpenCode 2 plugin API replaces the V1 config hook, so the shipped
plugin no longer loads on 2.x. Port every OpenCode surface in one change:
- Register all bundled skills and the three specialist subagents through
ctx.skill.transform and ctx.agent.transform, with ordered V2 permission
rulesets and frontmatter-stripped skill content.
- Move the shared definitions helper to .opencode/lib so the plugin
loader does not discover it as a plugin entrypoint.
- Rewrite the behavioral eval runner around opencode serve, the V2
session API, and skill attachments; keep the grading, redaction, and
snapshot evidence model unchanged.
- Port the smoke test to @opencode/cli 2.0.14 and verify registration
through a companion plugin, and update CI to install dependencies.
- Require OpenCode >= 2.0.0 and depend on @opencode/plugin ^2.0.14.
Add a least-privilege HOL Guard workflow, scanner policy, security reporting guidance, dependency monitoring, and immutable GitHub Action references. Align the Codex marketplace with its local plugin source and keep scanner verification reproducible for the awesome-ai-plugins submission.
Add a complete README quick start for Codex, OpenCode, Claude Code, and Pi, explain capability profiles and platform requirements, preserve the shell-contained security boundary, and protect the documentation with regression coverage.
Rely on the required release CI before uploading to GitHub Packages so the immutable version 0.5.0 tag can be mirrored without inheriting registry configuration inside package tests.
Use the public npm registry for temporary consumer installations so release workflows can publish to GitHub Packages without redirecting test dependencies away from npmjs.
Align npm and host plugin metadata, refresh the release workflow example, update the branded version marker, and record the final sandbox compatibility verification date.
Create upstream proxy sockets, generated certificates, and temporary files inside each command's disposable root, and disable mutable host system Git configuration without broadening filesystem access.
Enable correlated macOS Seatbelt diagnostics for failed commands so runtime denials identify the exact blocked operation while preserving exit codes and cleanup. Update the Linux shared temporary-path integration assertion to verify its isolated tmpfs behavior and confirm that no host file is created.
Add the missing Seatbelt interpreter permission through an exact-version, fail-closed compatibility shim so shell and env shebang commands work without weakening filesystem visibility. Align the Linux credential-directory test with its disposable tmpfs behavior and document when the temporary upstream compatibility layer can be removed.
Allow bounded parent directories for symlinked interpreters required by Bubblewrap, expose the macOS shell selector, and reject executable paths that overlap credential directories even through filesystem aliases. Add unit and real containment regressions for these platform-specific cases.
Resolve and expose the pinned seccomp helper before Linux execution, make doctor validate the same platform resources, filter invalid host environment keys from backend declarations, and canonicalize cross-platform path assertions so real containment checks behave consistently on GitHub runners.
Pin Anthropic's sandbox runtime and translate trusted skill profiles into lightweight, on-demand filesystem, network, credential, and process-tree restrictions. Add readiness diagnostics, secure command and Git workspace discovery, signal-safe cleanup, Linux and macOS integration coverage, packaged-runtime checks, and practical compatibility guidance for Codex, Claude Code, OpenCode, and Pi.
Define versioned capability profiles for every bundled skill, add policy resolution and diagnostic commands, preserve sandbox metadata during selective installation, and cover the contract and packaged artifacts with tests.