The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes local project files to identify the runtime environment, creating a surface for indirect prompt injection.
Ingestion points: SKILL.md (Step 0) identifies files like package.json, pyproject.toml, requirements.txt, go.mod, Cargo.toml, pom.xml, and Gemfile.
Boundary markers: The instructions do not specify delimiters or provide "ignore embedded instructions" warnings for the agent when processing these external files.
Capability inventory: The skill is configured with access to powerful tools including Bash, Write, Edit, WebFetch, and WebSearch (SKILL.md frontmatter).
Sanitization: No sanitization or validation steps are described for the content read from project files before the agent processes them to generate configurations.
[METADATA_POISONING]: The skill contains inconsistent metadata across its files, which could lead to confusion regarding the current version and capabilities.
Evidence: SKILL.md lists version 0.6.4, while .claude-plugin/plugin.json lists version 1.9.1.

relay-docker