The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from GitHub issues which could potentially contain malicious instructions.
Ingestion points: Reads titles, bodies, and comments from GitHub using gh issue view in SKILL.md.
Boundary markers: Absent; there are no instructions to the agent to treat the issue data as untrusted or to ignore embedded commands.
Capability inventory: The agent can execute shell commands (curl, gh, jq), use MCP tools (veris), and run local repository scripts (.veris/run.sh).
Sanitization: Absent; the skill processes issue content directly to identify load-bearing claims.
[COMMAND_EXECUTION]: The skill executes various CLI tools and repository-local scripts as part of its workflow.
Evidence: Uses gh, curl, and jq for data processing, and executes .veris/run.sh for integration testing.
[EXTERNAL_DOWNLOADS]: The skill retrieves data and configuration from external vendor infrastructure.
Evidence: Interacts with https://svc.api.veris.ai for sandbox orchestration and schema retrieval.

build