The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses metadata and instructions that involve executing Babashka (Clojure) and shell commands for environment validation and data navigation.
Evidence: The probe field in the frontmatter contains bb -e '(println ...)' and the usage section provides grep, ls, and wc commands for querying local files.
Context: These commands are used legitimately for project-specific research navigation and sanity checks.
[DATA_EXFILTRATION]: The skill instructions direct the agent to access and read multiple files from the user's home directory (~/).
Evidence: References to files such as ~/amp-thread-*.md, ~/relational-thinking-skill-structure.md, and the ~/ies/ directory in SKILL.md.
Context: These paths are used to locate and organize specific research threads within the author's ecosystem.
[PROMPT_INJECTION]: The skill is designed to ingest and process a large number of external markdown files, which presents a surface for indirect prompt injection.
Ingestion points: The agent is instructed to read content from various markdown threads located in ~/ (e.g., amp-thread-*.md).
Boundary markers: There are no instructions provided to treat the content of these threads as data rather than instructions or to use delimiters to prevent accidental command obedience.
Capability inventory: Across the scripts and usage guides, the agent has access to shell execution, SQL execution via DuckDB, and Babashka scripting.
Sanitization: No evidence of validation or sanitization of the external file content is present.

relational-thinking