The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on shell command execution for its core functionality.
Evidence includes the use of uv run to execute Python modules (src.integrations.snaptrade.sync_transactions_db, src.integrations.simplefin.sync_expenses_db, src.integrations.refresh_all).
Direct execution of sqlite3 commands to query and manipulate the family_office.db database.
Use of complex awk and sort pipelines in workflows/IngestTransactions.md to parse and analyze CSV files from the user's ~/Downloads directory.
Execution of inline Python scripts via uv run python -c to audit coverage and detect stale connections.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external financial feeds (SnapTrade, SimpleFIN) and local CSV exports (Fidelity) which are then presented to the agent for summarization and review.
Ingestion points: Financial transaction descriptions and memos are read from family_office.db and ~/Downloads/History_for_Account_{account_id}.csv (documented in SKILL.md and IngestTransactions.md).
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when the agent processes transaction descriptions for 'Spending reviews' or 'Sync Summary' generation.
Capability inventory: The agent has capabilities including shell command execution (uv run, sqlite3, awk), file system operations (cp, ls), and access to environment variables via .env files.
Sanitization: There is no evidence of sanitization or filtering of transaction descriptions before they are incorporated into the agent's context for reporting.

transaction-syncing