The Agent Skills Directory

[SAFE]: The skill is authored by 'posthog' and is designed to audit its own platform entities (experiments and feature flags). All identified logic and referenced checks in 'references/experiment-checks.md' and 'references/flag-checks.md' align with the stated purpose of improving project hygiene and configuration correctness.
[COMMAND_EXECUTION]: No unauthorized or dangerous shell commands were detected. The skill instructions in 'SKILL.md' focus on using platform-specific tools (read_data, list_data, create_notebook) for reporting and data retrieval.
[DATA_EXFILTRATION]: No patterns of data exfiltration were found. The skill does not access sensitive system files (e.g., SSH keys, environment secrets) or make external network requests to non-whitelisted domains.
[PROMPT_INJECTION]: The skill does not contain instructions aimed at bypassing agent safety filters or overriding core behavioral constraints. Instructions are focused on the technical audit process.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns, such as piping curl to bash or downloading untrusted scripts, were identified across any of the referenced files.
[SAFE]: Potential for indirect prompt injection was evaluated and found to be minimal. Ingestion points: Untrusted data enters the agent context via read_data and list_data calls for experiments and feature flags (specifically name and description fields) as described in 'SKILL.md' and 'references/experiment-checks.md'. Boundary markers: The instructions do not explicitly mandate the use of delimiters or 'ignore' instructions for the processed data. Capability inventory: The skill uses read-only tools and create_notebook for reporting; no destructive or system-level write capabilities are utilized. Sanitization: No explicit sanitization of external metadata is performed before reporting.

auditing-experiments-flags