The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to fetch sensitive financial information (transactions, accounts, and categories) from the Toshl API and store it in local CSV and JSON files within the toshl/data/ directory. The instructions explicitly state that this cache is "git-tracked". Storing unencrypted financial records in a version-controlled directory poses a significant risk of data exposure or exfiltration if the repository is pushed to a remote host (e.g., GitHub) or shared with unauthorized users.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection. It ingests untrusted data from the Toshl API (specifically transaction descriptions and tag names) and processes it to generate reports. An attacker could potentially embed malicious instructions within transaction data that could influence the agent's behavior during the reporting or synchronization phase.
Ingestion points: get_entries tool output from the Toshl MCP server, subsequently read from data/{YYYY-MM}_entries.csv.
Boundary markers: None identified; the skill processes raw CSV data and interpolates it directly into Markdown templates.
Capability inventory: Write (file creation), Read (data processing), Bash (workflow execution), and Edit.
Sanitization: No evidence of sanitization or validation of external transaction data before report generation.

toshl