atmos-introspection
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides a detailed reference for executing the
atmosCLI tool, documenting various subcommands for workspace introspection, stack management, and configuration analysis.\n- [EXTERNAL_DOWNLOADS]: The--docs=mdfeature allows the tool to fetch markdown documentation from remote URLs for rendering within the terminal.\n- [DATA_EXFILTRATION]: The skill documents the--uploadflag used to send configuration data or component instances to external HTTP endpoints or the Atmos Pro API as part of standard tool functionality.\n- [DYNAMIC_EXECUTION]: Theatmostool processes Go templates and custom YAML functions (e.g.,!terraform.output) when resolving stack configurations, enabling dynamic logic within the ingested files.\n- [INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection because the agent is instructed to ingest and process data from local workspace configurations and remote documentation.\n - Ingestion points: Local Atmos stack manifests and remote markdown documentation files fetched at runtime.\n
- Boundary markers: No specific instructions or delimiters are provided to the agent to prevent it from following instructions that might be embedded in the processed configuration or documentation.\n
- Capability inventory: The agent uses the tool to read, merge, and filter (
yq) configuration data which can contain dynamic logic via templates and functions.\n - Sanitization: The skill does not specify any sanitization or validation of the ingested content before it is processed and returned to the agent context.
Audit Metadata