The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from plugin configuration files to perform release actions. Ingestion points: Manifest and configuration files including package.json, pnpm-lock.yaml, cordis.patch.yml, and dsh.plugin.json as specified in SKILL.md. Boundary markers: The instructions lack explicit boundary markers or delimiters to prevent the agent from interpreting data within these files as instructions. Capability inventory: The skill utilizes powerful capabilities including shell command execution (npm, pnpm, git, node), network access for publishing (npm publish), and file system modifications in references/publish-playbook.md and SKILL.md. Sanitization: Beyond standard semantic versioning checks in scripts/verify-release.mjs, there is no mention of sanitizing or validating the content of ingested manifests.
[COMMAND_EXECUTION]: The skill executes build, package management, and version control commands as part of the release process. Evidence: npm pack, pnpm install, pnpm run build, git push, and git clone commands are used in SKILL.md and reference playbooks.
[EXTERNAL_DOWNLOADS]: The skill fetches code and dependencies from external repositories and registries. Evidence: The skill downloads source code from github.com/deepseek-ai/deepseek-harness.git and utilizes the npm registry for package publication and validation.
[DYNAMIC_EXECUTION]: The skill uses Node.js to evaluate version strings and execute validation logic dynamically. Evidence: Inline scripts using node -p and node -e to parse package.json and perform semver comparisons are found in references/publish-playbook.md.

plugin-release