ScienceClaw: Local File Investigation

Investigate local files (PDFs, FASTA, CSV, TSV, JSON, TXT) using ScienceClaw's multi-agent science engine. Accepts files shared in chat or paths on disk, ext...

This skill is internally consistent with its description, but it runs a local Python program (bin/scienceclaw-post) from a directory you must already have on disk and will likely send file contents and workspace context to an LLM (Anthropic) and external science services (BLAST, UniProt, PubChem, PubMed, PDB, etc.). Before installing/using: 1) Inspect the SCIENCECLAW_DIR and the bin/scienceclaw-post script (and the .venv) so you understand what code will run locally. 2) Confirm you are comfortable with the file types being uploaded/shared with external services and an LLM — do not use this on sensitive data or regulated biological sequences without review. 3) If you lack the local ScienceClaw install, the instructions will fail rather than silently download code — prefer to run it in a sandboxed environment first. 4) Consider removing or rotating ANTHROPIC_API_KEY if you only want local, offline analysis. If you want a higher-assurance recommendation, provide the contents of bin/scienceclaw-post (or the project's README) so its network calls and data handling can be inspected.

Type: OpenClaw Skill Name: scienceclaw-local-files Version: 1.0.2 The skill bundle provides instructions for an AI agent to execute a local Python script (`scienceclaw-post`) using shell commands constructed from user-provided inputs like `FILE_PATH` and `TOPIC`. This pattern introduces a high risk of shell injection vulnerabilities if the agent does not properly sanitize these variables before execution. While the stated purpose is scientific data analysis and the logic appears aligned with that goal, the reliance on shell execution with variable arguments in `SKILL.md` is a significant security risk.