pp-lightroom-classic
Warn
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of a command-line interface (CLI) from external sources. It provides instructions to use
npx -y @mvanhorn/printing-press-libraryandgo install github.com/mvanhorn/printing-press-library/...to download and install the binary. - [COMMAND_EXECUTION]: The skill implements an automated 'learning loop' (Steps 1-6) that journals every invocation into a local database and journal file (
teach.log). The agent is explicitly instructed to execute 'trial commands' from generated 'candidates' verbatim. This mechanism allows the execution of commands derived from historical data, which could be manipulated. - [DATA_EXFILTRATION]: Every command in the CLI supports a
--deliver webhook:<url>flag. This enables the agent to send command output—which contains information from the local Lightroom catalog including file paths, camera details, and EXIF metadata—directly to an arbitrary external URL via POST requests. - [PROMPT_INJECTION]: The
recallandteachpipeline creates an indirect prompt injection surface. Data encountered in processed catalogs or previous user queries is stored in the learning journal and used to auto-synthesize future 'playbooks' and 'notes', which could influence or override the agent's subsequent logic.
Audit Metadata