The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data originating from external application logs and error occurrences, which could potentially contain malicious instructions intended to influence the agent.
Ingestion points: Data enters the context via posthog:query-logs, posthog:execute-sql (reading from events and logs tables), and posthog:query-error-tracking-issue-events (specifically in SKILL.md and referenced files).
Boundary markers: The skill lacks explicit structural boundary markers or "ignore instructions" warnings to separate log content from agent instructions, though it does include a privacy instruction to "Exclude secrets and unneeded personal data from log excerpts."
Capability inventory: The agent has access to powerful tools including posthog:execute-sql, which allows direct database interaction, and various API tools for retrieving traces and session recordings.
Sanitization: The instructions mitigate risks by directing the agent to use "validated UUID" and "validated hex ID" literals within the SQL templates provided in references/finding-exceptions.md and references/finding-trace-logs.md.
[COMMAND_EXECUTION]: The skill involves the dynamic generation and execution of SQL queries against a database environment.
The files references/finding-exceptions.md and references/finding-trace-logs.md contain SQL templates that the agent must populate with IDs extracted from user input or previous tool outputs.
The instructions provide specific guidance on using hex conversion (unhex) and base64 encoding (base64Encode) within queries to handle trace IDs correctly and safely.

linking-errors-to-traces