hugging-face-dataset-viewer

Warn

Audited by Gen Agent Trust Hub on Aug 9, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXPOSURE]: The skill accesses highly sensitive local directories containing agent interaction history, project details, and potentially secrets, specifically ~/.claude/projects, ~/.codex/sessions, and ~/.pi/agent/sessions.
  • [DATA_EXFILTRATION]: Provides explicit instructions to transmit local agent session traces to Hugging Face dataset repositories, which poses a privacy risk if traces contain personally identifiable information (PII) or credentials.
  • [EXTERNAL_DOWNLOADS]: Downloads and executes the official @huggingface/hub package from the NPM registry via npx to facilitate data uploads.
  • [COMMAND_EXECUTION]: Spawns shell commands to perform repository management and data transfers using curl and the hf CLI.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external dataset APIs, creating a potential attack surface for instructions embedded in those datasets to influence the agent's behavior.
  • Ingestion points: Data retrieved from API endpoints such as /rows, /search, and /filter in SKILL.md.
  • Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands in the processed datasets.
  • Capability inventory: Includes shell execution (npx, curl), file system access (~/.codex/sessions), and repository management tools.
  • Sanitization: No sanitization or validation logic is specified for the content retrieved from external datasets.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 9, 2026, 09:27 PM
Security Audit — agent-trust-hub — hugging-face-dataset-viewer