The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes untrusted merchant descriptions from external files.
Ingestion points: According to SKILL.md, the skill ingests data from CSV bank exports and user-pasted transaction text.
Boundary markers: No delimiters or safety instructions are provided to isolate transaction descriptions from the agent's logic.
Capability inventory: The skill is capable of writing HTML and JSON files to the local filesystem and executing CLI commands.
Sanitization: The workflow specifies normalization of numeric values but does not describe any sanitization for string-based merchant descriptions to prevent them from influencing agent behavior or injecting malicious HTML.
[COMMAND_EXECUTION]: The skill's operational model depends on local shell command execution.
Evidence: Multiple instructions in SKILL.md and references/budget-templates.md require the agent to invoke python -m watch_my_money for analysis, comparisons, and budget management.
Context: These commands utilize the vendor's own Python module ('sundial-org') to perform the core functions of the skill.
[EXTERNAL_DOWNLOADS]: The generated report references resources from a well-known service.
Evidence: The assets/template.html file fetches font styles from Google's official repository at fonts.googleapis.com.

watch-my-money