skills/dotlas/skills/look-into-it/Gen Agent Trust Hub

look-into-it

Pass

Audited by Gen Agent Trust Hub on Jul 24, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data from external chat threads (Slack) to define its investigation tasks and fixing logic.
  • Ingestion points: External Slack thread content accessed via slack_read_thread in the Digest step.
  • Boundary markers: Absent; there are no specified delimiters to separate untrusted chat content from agent instructions.
  • Capability inventory: File system read/write/delete (writing diagnostic-<slug>.md plans and 'Implement' step), Slack tool access (slack_read_thread, slack_add_reaction), and execution of the /loop command.
  • Sanitization: Not specified for external content, though the skill requires human interaction (AskUserQuestion) before running code experiments.
  • [DATA_EXFILTRATION]: The skill reads codebase materials, including identifiers such as file/table names and error strings, and is instructed to post 'milestone updates' back to a Slack thread. This facilitates the movement of technical codebase information to an external communication channel.
  • [COMMAND_EXECUTION]: The skill invokes the /loop command for periodic update checks and involves modifying the local filesystem to create implementation plans and apply fixes.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 24, 2026, 11:31 AM
Security Audit — agent-trust-hub — look-into-it