$ the-wire · showcase
Claude Code Action closes a shell injection path in structured output
By RepoJournal · Filed · About Anthropic · Composed from the cited sources · methodology
The day's real work is narrow: claude-code-action stopped writing model output directly into run: scripts, while four NEAR repositories opted in to GitHub security advisories and plugin-dev moved its MCP examples onto local server files.
Pass structured output to shell steps through env anthropics/claude-code-action
GitHub interpolates a ${{ }} value into a run: script before the shell ever reads it, so a quote in Claude's answer could break the script or execute as a command. structured_output now reaches shell steps through env:, matching the pattern already used for head_bran, and a test asserts the example, docs, and test workflow keep doing so.
plugin-dev: use local servers in MCP examples anthropics/claude-plugins-official
Two examples in the MCP server types reference now run a server file inside the plugin rather than pointing at an external endpoint. plugin-dev is bumped to 0.1.2.
nearcore: opt in to GitHub security advisories anthropics/oss-scanner
The same opt-in landed across nearcore, borsh-rs, and near-sdk-rs, putting all three under GitHub's advisory reporting flow for the Rust NEAR stack.
Merge pull request #1431 from anthropics/mhegazy/bump-security-guidance-2.0.19 anthropics/knowledge-work-plugins
The security-guidance plugin moved from c07c748f to dceb15d5 in the knowledge-work-plugins repo.
chore: Update mods/ anthropics/claude-code
by GitHub Actions (publish-mods)
Everything else was housekeeping: an automated publish-mods refresh of mods/ in claude-code, plus the merges that carried the advisory opt-ins and the security-guidance bump.