debugging-experiments
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides a standalone Python script,
srm_check.py, designed to be run locally by an operator. The instructions guide the user to execute this script usingpython3to analyze CSV data exported from PostHog queries. This is an expected diagnostic function for staff and does not involve automated or remote execution. - [DYNAMIC_EXECUTION]: The script
srm_check.pyimplements a re-implementation of PostHog's internal hashing algorithm (using SHA1) to verify experiment assignment. While it useshashlib.sha1, the script includes explicit warnings that this is for compatibility with the product's bucketing logic rather than security, and it includes a self-test mode to ensure mathematical correctness. - [INDIRECT_PROMPT_INJECTION]: The workflow involves parsing support tickets, which are external, untrusted inputs. The skill addresses this by explicitly instructing the agent to treat ticket text and query results solely as data and evidence, never as instructions to override behavior or bypass safety filters. It includes a specific warning to stop and quote the content to an operator if it appears to be an injection attempt.
- [SAFE]: All external references (US and EU PostHog domains), tools (PostHog MCP tools), and organizational links belong to the vendor's own infrastructure. The use of staff impersonation and Django admin is scoped to official support procedures for established customers.
Audit Metadata