The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/skill_router.py scans multiple local directories, including project-specific paths (.opencode/skills) and user-level configuration paths (~/.config/opencode/skills), to collect the names and descriptions of all installed skills. This metadata is then transmitted to an external service via the typesafe-sdk API to perform task routing.
[INDIRECT_PROMPT_INJECTION]: The skill lacks sufficient boundary markers when processing user-supplied queries, creating a surface for indirect prompt injection attacks where a malicious query could influence the routing logic or model behavior.
Ingestion points: The --query command-line argument in scripts/skill_router.py is passed as the task parameter to the system_one model call.
Boundary markers: None. The query is interpolated directly into the model's state without delimiters or instructions to treat the content as untrusted data.
Capability inventory: The script possesses the capability to read and output the full text of any SKILL.md file found in the scanned directories if the --load flag is provided.
Sanitization: No validation or sanitization is performed on the query string before it is sent to the routing model.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the typesafe-sdk and pyyaml libraries. These dependencies are fetched from external package registries to enable the skill's core routing and YAML parsing functionality.

jev-skill-router