hindsight-cloud

Fail

Audited by Gen Agent Trust Hub on Mar 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill documentation includes a command to download and execute a shell script directly from the vendor's domain (https://hindsight.vectorize.io/get-cli | bash) to install the Hindsight CLI tool.\n- [DATA_EXFILTRATION]: The skill reads and writes a configuration file at ~/.hindsight/config containing a user-provided API key. It communicates with the vendor's API endpoint (api.hindsight.vectorize.io) to store and retrieve data.\n- [COMMAND_EXECUTION]: The skill executes shell commands to check configuration, create directories, and run the hindsight CLI for memory operations.\n- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by design.\n
  • Ingestion points: Knowledge retrieved from a shared, team-accessible memory bank via hindsight memory recall is injected into the agent's context (SKILL.md).\n
  • Boundary markers: None. Recalled context is provided to the agent without delimiters or instructions to ignore embedded commands.\n
  • Capability inventory: The agent has the ability to execute shell commands, read/write local files, and perform network operations via the hindsight CLI (SKILL.md).\n
  • Sanitization: No validation or escaping is applied to the data retrieved from the remote memory bank before it is processed by the agent.
Recommendations
  • HIGH: Downloads and executes remote code from: https://hindsight.vectorize.io/get-cli - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 16, 2026, 09:13 PM