The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to "operate autonomously and proactively" and to refine code "without requiring explicit requests." This reduction in human-in-the-loop oversight is a risk factor that may allow the agent to perform unintended modifications if triggered by malicious context.
[PROMPT_INJECTION]: The skill processes "recently modified code" as an ingestion point for untrusted data without establishing protective boundaries or sanitization steps, leaving it vulnerable to indirect prompt injection. 1. Ingestion points: recently modified source code files. 2. Boundary markers: absent; there are no instructions to ignore or delimit embedded commands within analyzed files. 3. Capability inventory: the agent is expected to read, analyze, and write code to the filesystem. 4. Sanitization: absent; the instructions do not include validation or filtering of the code being processed.

simplify