The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize untrusted data from the local project environment, including READMEs, manifests (package.json, Cargo.toml), and source code. This content is aggregated into a learning document and then passed as input to the /infographic tool. A malicious project file containing hidden instructions could potentially influence the agent's behavior or the output of the downstream tool.
Ingestion points: Local project files, including configuration manifests, source code, and Git history.
Boundary markers: None identified; the skill does not wrap the ingested project content in delimiters or include warnings to ignore embedded instructions.
Capability inventory: The skill performs extensive file reading, writes to the local file system (tmp/learn.md), and executes the /infographic tool.
Sanitization: None identified; project content is used directly for summarization and tool input.
[COMMAND_EXECUTION]: The instructions direct the agent to remind the user to execute the make claude shell command. While this is a recommendation to the user rather than direct execution by the agent, it represents a social engineering surface where a skill could encourage users to run arbitrary local commands.

learn