Shopify Changelog now connects announcements to the project being developed and can supply that evidence automatically through agent hooks.
New
inspect: bounded local discovery of Shopify packages, selected source identifiers, extension targets, and scoped API version evidence, with file/line locations.check: task-focused release candidates with source links, publication dates, release conditions, and cautious version comparisons. Webhook versions are kept separate from API client versions. Identifier matches are prioritized over surface-only fallback results.check --offline: cached checks with explicit freshness and coverage. Incomplete evidence exits with code 2; it is not reported as a clean compatibility result.setup.py --hooks: optional offline SessionStart and UserPromptSubmit hooks for Codex and Claude Code, preserving unrelated hooks/settings. Hooks are quiet for unrelated prompts and non-Shopify projects and never block prompts.doctor: runtime and hook configuration diagnostics.- Shared-install detection avoids duplicate Codex skill entries when using the Skills CLI.
Install
git clone https://github.com/buildwithzach/shopify-changelog.git
cd shopify-changelog
python3 install.py --agent both --auto-reference --hooks
~/.local/bin/shopify-updates syncOr install just the skill:
npx skills add buildwithzach/shopify-changelog --skill shopify-changelog -g -a codex claude-codeFor existing clone installs, pull the update and rerun setup with the desired options. For Skills CLI installs, update the skill and ask your agent to enable its optional hooks. New Codex hooks require review/trust through /hooks; this release does not bypass trust or disabled-hook policies.
Validation and limits
35 automated tests pass. A fresh install from the public repository passed project matching, version-scope, and hook-contract smoke tests against the live archive. Configured hook commands completed successfully in direct tests. Codex CLI 0.144.5 reports hook support enabled; Claude Code 2.1.211 launches successfully. Native host-driven hook execution after trust/restart is not claimed.
Matching is a bounded lexical heuristic, not a compatibility certification or an accuracy benchmark across production projects. Hooks are offline and ask the agent to refresh stale data before relying on it. They do not upload project code, log prompts, or execute repository scripts. See the README and engineering notes for scanning and feed-history limits.