anfra-convert-looker
Warn
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/lib/sources.mjsutilizeschild_process.execFileSyncto executegitcommands (such asrev-parse,diff, andls-files) to retrieve repository metadata and commit hashes. While this is used for an audit trail, spawning subprocesses with system tools is a sensitive operation. - [DYNAMIC_EXECUTION]:
- The script
scripts/lookml-parser/transformations/assemble-models.jsdynamically generates regular expressions usingnew RegExp()based on LookML include patterns found in user files. - The utility
scripts/lookml-parser/maybe-yaml-parse/index.jsperforms dynamic loading of theyamllibrary using theimport()statement. - [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where a second pass is performed by an AI agent (subagent) that reads generated AML files. These files contain reprinted LookML snippets from user-provided files within
TODOandREVIEWcomments. A maliciously crafted LookML file could attempt to inject instructions to manipulate the subagent's logic during this phase. - Ingestion points: LookML files (
.lkml) are read from the filesystem byscripts/convert.mjs. - Boundary markers: User-provided SQL and LookML snippets are wrapped inside
// TODO(lookml-to-aml:...)or// REVIEW(lookml-to-aml:...)blocks. - Capability inventory: The skill has read/write access to the local filesystem and the ability to execute shell commands via
execFileSync. - Sanitization: LookML content is reprinted from a parsed AST, providing basic structural normalization, but no semantic sanitization is applied to the content placed in comments.
Audit Metadata