The Agent Skills Directory

[DATA_EXFILTRATION] (MEDIUM): The scripts scripts/analyze_all_skills.clj, scripts/print_every_skill.clj, and scripts/random_walk_verifier.clj perform broad directory traversal on the path ~/.claude/skills.
These scripts read and extract information (descriptions and full content) from every SKILL.md file found in other skill directories.
This allows the skill to expose the user's entire local skill ecosystem, including proprietary logic or prompts from other tools, to the agent context.
[COMMAND_EXECUTION] (LOW): The skill relies on the execution of Babashka (bb) scripts for its core functionality. While this is the intended design, the meta-analysis scripts exercise filesystem access that extends beyond the skill's own installation path.
[EXTERNAL_DOWNLOADS] (SAFE): The skill fetches biological data from https://www.ebi.ac.uk/ols4/api/. This is a legitimate and expected scientific resource for the skill's stated purpose of ontology translation.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill's meta-analysis feature ingests untrusted data from other skills.
Ingestion points: scripts/analyze_all_skills.clj and scripts/random_walk_verifier.clj read SKILL.md files from other directories.
Boundary markers: None identified; external content is processed directly via regex.
Capability inventory: File system read access and network GET requests via babashka.http-client.
Sanitization: None; the scripts extract text directly from the external skill files.

yb-translator