The Agent Skills Directory

[SAFE]: The instructions in SKILL.md follow best practices for persona definition and task execution. There are no attempts to bypass safety filters or override agent behavior.
[SAFE]: The shell script in scripts/run.sh serves as a standard entry point for the skill. It correctly handles input via command-line arguments or stdin and delegates execution to a project-level runner script without performing any unauthorized actions.
[PROMPT_INJECTION]: The skill inherently processes untrusted data (code for review), which constitutes a surface for indirect prompt injection. However, the risk is negligible as the skill lacks high-privilege capabilities such as direct file system writes or network access.
Ingestion points: The skill ingests code through the --stdin flag or command-line arguments in scripts/run.sh.
Boundary markers: Absent. The prompt does not explicitly instruct the agent to ignore instructions embedded within the code snippets being reviewed.
Capability inventory: The skill itself has no tool definitions or subprocess capabilities. It relies entirely on the host environment's run-skill.py (external to this skill) for execution.
Sanitization: Absent; input is passed directly to the model context.

engineering-code-reviewer