[svg153]

Summary

Publishes the Agent Plugins 1.0 pilot as v1.2.3 after completing current-client compatibility evidence and release hardening.

This is a patch release: the canonical Build-or-Reuse decision behavior is unchanged.

Included since v1.2.2

• Copilot CLI 1.0.83 marketplace install/discovery evidence;
• Codex CLI 0.153.4 native Agent Plugin install/discovery evidence;
• Claude Code 2.1.263 strict validation + marketplace install/discovery evidence;
• Gemini CLI 0.58.0 extension validation/install/discovery evidence;
• GitHub Agentic Workflows 0.88.4 compile/consumption evidence with immutable release-ref resolution;
• Claude adapter generation hardened to omit host-irrelevant interface metadata;
• immutable refs/tags/v* ruleset recipe + validator/admin helper (server-side activation remains tracked in security: protect release tags with an immutable v* ruleset #32);
• evidence-based compatibility matrix and separated behavioral/update-parity follow-up runtime: verify behavioral and update parity before retiring adapters #35.

Version contract

• metadata.yaml: 1.2.3
• canonical SKILL.md: 1.2.3
• portable Agent Plugin + versioned generated adapters: 1.2.3
• CHANGELOG.md: 1.2.3 dated 2026-09-07
• release notes: docs/releases/v1.2.3.md

Generated manifests are derived from the same canonical state; no runtime skill copy is introduced.

Release automation

After merge to main, the existing Release workflow should create v1.2.3 targeting the merge commit because the version does not yet exist.

The pilot issue will be closed only after the GitHub Release is observed successfully, not merely on PR merge.

Traceability

Refs #28
Behavior/update parity follow-up: #35
Release-tag server-side hardening: #32
Parent initiative: svg153/skills#33

[svg153]

Lifecycle follow-up note: current Copilot CLI exposes native plugin update/uninstall, while current Codex CLI exposes plugin add/list/remove plus marketplace upgrade but no plugin-update command. This means Codex remains a retirement candidate at most; update/governance parity is not yet equivalent. Track the stronger test in issue #35 rather than treating install/discovery as adapter-retirement evidence.