The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitHub issues and user-provided prompts to define development tasks and execution flows, creating an attack surface for indirect instructions.
Ingestion points: GitHub issue content (title, body, and comments) accessed via the gh issue view command, and free-text user prompts.
Boundary markers: The skill instructs the agent to "Quote" the request, providing a basic delimiter between the instructions and the external data.
Capability inventory: The skill uses the veris CLI for environment management and code execution (veris run), the gh CLI for repository interaction, and local file system access for configuration storage in the .veris/ directory.
Sanitization: No specific sanitization or filtering logic is described for the ingested GitHub or prompt content.
[COMMAND_EXECUTION]: The skill relies on standard command-line interfaces to interact with the repository, sandbox environments, and issue trackers.
Evidence: The workflow involves frequent use of gh for fetching issues and veris for lifecycle management (up, status, sandbox, run, down).
[DYNAMIC_EXECUTION]: The skill dynamically executes application code and user-defined flows as part of its core build and verification process.
Evidence: The veris run command is used to execute the repository's code path within a sandbox, involving mounting the local working tree into a runtime container or environment.

build