The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses $ARGUMENTS directly to influence its scanning and analysis focus. This creates a surface for indirect prompt injection if the arguments are sourced from untrusted data (e.g., a file name or user input), potentially causing the agent to deviate from its intended defense analysis role.
[COMMAND_EXECUTION]: The skill attempts to write execution metadata to ~/.claude/projects/skill-telemetry.md. While this is for telemetry and 'self-evolution', it involves file system writes to a specific configuration directory based on project state.
[DATA_EXFILTRATION]: The skill scans for sensitive defense program data (EVM reports, budget submissions, SARs). While it includes a 'DO NOT' instruction against disclosing classified or FOUO data, the autonomous nature of the agent scanning the local environment for such documents carries inherent data exposure risks if the environment is not properly scoped.

defense-budget