The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by instructing the agent to read and analyze local log files, such as ./logs/sys-console.log. If these logs contain data from external users (e.g., logged request parameters or usernames), an attacker could embed malicious instructions that the agent might inadvertently follow during troubleshooting.
Ingestion points: The agent is instructed to read local log files in both SKILL.md and references/error-patterns.md.
Boundary markers: No explicit delimiters or instructions are provided to distinguish between system logs and untrusted data within those logs.
Capability inventory: The agent has the capability to read local files and execute shell commands for filtering and searching.
Sanitization: There is no mention of sanitizing or escaping the content of log files before analysis.
[COMMAND_EXECUTION]: The skill suggests the use of standard Unix utilities and network tools to assist in diagnostics. While these are used for their intended purpose in a developer context, they represent a functional capability for command execution.
Evidence: SKILL.md includes templates for grep, sed, awk, tail, and curl for log searching and API testing.

bug-detective