The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process third-party plugin directories, including manifests, instructions, and executable code, which creates a surface for indirect prompt injection attacks.
Ingestion points: The workflow involves locating and inventorying the plugin root, manifest files, and component source code as described in SKILL.md (Step 8) and references/requirements.md.
Boundary markers: Absent; the skill instructions do not provide explicit delimiters or instructions to the agent to isolate or disregard potential instructions contained within the candidate plugin files.
Capability inventory: The agent uses file system tools to read and inventory files and has the capability to execute commands like npx for validation.
Sanitization: Absent; there are no instructions for escaping or sanitizing the content of the external files before they are processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill recommends downloading and running the official Anthropic Claude Code validator using a specific pinned version.
Evidence: npx --yes @anthropic-ai/claude-code@2.1.287 plugin validate ./candidate-plugin --strict in SKILL.md.
Note: This targets a well-known service (Anthropic) using a pinned version to ensure structural integrity of the plugin.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command to validate the plugin package.
Evidence: The execution of the npx command as described in Step 9 of the "Check the exact native package" section in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill references a structural checker script hosted on a GitHub repository belonging to the skill author.
Evidence: https://github.com/khadinakbarlabs/expo-mobile-app-builder/blob/main/scripts/check-anthropic-package.mjs in SKILL.md.
Note: This represents a vendor-owned resource provided by the author (khadinakbarlabs).

prepare-anthropic-plugin