The Agent Skills Directory

[SAFE]: The skill performs all data processing using a local Python engine (scripts/analyse.py), ensuring that sensitive calculations and data manipulation occur entirely within the user's environment without external network calls.\n- [SAFE]: The documentation includes explicit privacy disclosures, informing the user that while calculations are local, any data read into the agent's context is sent to the AI provider, which is standard behavior for LLM-based tools.\n- [SAFE]: The skill strictly follows a 'compute-then-interpret' methodology, utilizing deterministic scripts with precise numeric handling (Decimal) to prevent the AI from hallucinating figures or fabricating insights.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests external datasets (CSV, XLSX) provided by the user. 1. Ingestion points: Ingestion occurs via the ingest.read_any and ingest.read_large methods referenced in SKILL.md. 2. Boundary markers: The 'compute-then-interpret' workflow and explicit 'intent-gathering' steps (in SKILL.md) serve as structural boundaries to mitigate the risk of the agent following instructions embedded in data. 3. Capability inventory: The skill is limited to local file reading and writing (XLSX via openpyxl), with no network operations or subprocess calls present in the analysis engine. 4. Sanitization: Input data is parsed through standardized dataclean parsers for numbers, currencies, and dates.\n- [SAFE]: No evidence of obfuscation, persistence mechanisms, privilege escalation, or hardcoded credentials was found in the instructions or the analysis scripts.

data-analyse