The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/cross-review.sh executes standard git and text processing utilities (git, grep, sed, awk, wc) to calculate metrics like lines of code and file counts for PRs. It also facilitates the optional execution of the codex CLI tool for cross-vendor model evaluation.
[PROMPT_INJECTION]: The skill is designed to process untrusted external data as part of its core review function.
Ingestion points: scripts/cross-review.sh reads git diff output, and SKILL.md defines a workflow for reading feature specifications via the --spec parameter.
Boundary markers: No explicit delimiters or 'ignore embedded instructions' warnings are added to the untrusted content before it is passed to the reviewer models.
Capability inventory: The skill has access to shell execution, external CLI tool invocation, and the ability to dispatch new sub-agent sessions.
Sanitization: Input content is passed to reviewer models without sanitization or escaping.
[PROMPT_INJECTION]: The skill includes a 'Reflexive Trigger Rule' section in SKILL.md containing strong behavioral instructions. These are intended to enforce the P20 governance protocol autonomously and do not attempt to bypass core safety filters or ethical guidelines.

cross-review