The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and systematically analyze untrusted external data, including Incident Response Plans, playbooks, and runbooks. There are no instructions defining boundary markers to delimit this data or warnings to ignore instructions embedded within the processed files. This creates a surface where an attacker could embed malicious instructions in documentation that the agent then obeys during the analysis phase.
[COMMAND_EXECUTION]: The skill contains a 'Self-Evolution Telemetry' section that executes automated file system operations. It searches for project paths within ~/.claude/projects/ and appends execution metadata (outcome, iterations, bottlenecks) to a file named skill-telemetry.md. While intended for performance tracking, this represents silent, automated modification of files in a hidden system directory outside the primary scope of the incident response analysis.

incident-response