The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The skill utilizes babashka (bb) to execute arbitrary Clojure code via the -e flag. Specifically, it runs fs/glob over the entire user home directory (System/getProperty "user.home") with a depth of up to 6.
[DATA_EXFILTRATION] (MEDIUM): The skill systematically maps and reads files within the user's home directory. It defines functions like load-gayzip-manifest that use slurp to read local file contents into the agent's context. While no external exfiltration URL is explicitly visible, the broad recursive scanning and reading of the ~ directory constitutes a significant data exposure risk.
[DYNAMIC_EXECUTION] (MEDIUM): The skill relies on runtime execution of Clojure strings and dynamic loading of file content from computed paths on the filesystem.
[INDIRECT_PROMPT_INJECTION] (LOW):
Ingestion points: Reads .topos manifests and .tree files from ~/ies/rio/gayzip/ and ~/ies/hatchery_repos/.
Boundary markers: Absent; no delimiters or instructions are provided to the agent to ignore potentially malicious content within these files.
Capability inventory: Filesystem globbing and reading (slurp) via Babashka, which could be used to process malicious instructions found in the indexed files.
Sanitization: No sanitization or validation of the file content is performed before it is loaded into the agent's context.

topos-unified