The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill configures allowed-tools with Bash(python:*) and Bash(.venv/bin/python:*). This grants the agent arbitrary Python execution capabilities rather than restricting it to specific scripts or hardened parameters, significantly increasing the risk of unauthorized code execution.\n- [PROMPT_INJECTION] (HIGH): Indirect Prompt Injection Risk (Category 8). The skill's primary function is to ingest and analyze Django Silk profiling data, which contains attacker-controlled strings such as URL paths, request headers, and SQL queries. If malicious instructions are embedded in these logs, the agent may execute them using its broad shell permissions.\n
Ingestion points: Django Silk profiling data (SQL, URLs, headers) accessed via scripts/silk_debug.py.\n
Boundary markers: Absent. The skill provides no instructions to the agent to treat profiling data as untrusted or to ignore embedded commands.\n
Capability inventory: Bash (arbitrary python execution), Read, Grep.\n
Sanitization: Absent/Unverifiable. There is no evidence that scripts/silk_debug.py or the skill's logic filters malicious payloads from the profiling output.

silk-debug