The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill automatically processes external content from a public documentation repository to create instruction sets for the agent, creating an ingestion path for untrusted data.
Ingestion points: Fetches content from https://docs.rs/{crate}/latest/{crate}/ via agent-browser or WebFetch based on dependencies extracted from the local Cargo.toml file.
Boundary markers: No boundary markers or defensive prompt constraints are specified when formatting or embedding the external documentation text into local files.
Capability inventory: The skill creates directories and writes files under ~/.claude/skills/, and uses local commands like cat, grep, mkdir, and rm.
Sanitization: Lacks filtering or validation of the fetched text before using it to generate SKILL.md files or reference documents.
[COMMAND_EXECUTION]: Executes standard local shell utilities including cat, grep, mkdir, and rm to parse dependency files, build directory trees, and maintain the local cache layout.

core-dynamic-skills