The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to download and install external tools using npx -y @mvanhorn/printing-press-library and go install github.com/mvanhorn/printing-press-library/.... These resources are hosted on the npm registry and GitHub respectively.
[REMOTE_CODE_EXECUTION]: The installation process involves executing code from external repositories (NPM and GitHub) associated with the author. This is a standard installation procedure for this toolset but involves third-party code execution.
[DATA_EXFILTRATION]: The CLI tool supports a --deliver webhook:<url> flag, which allows the agent to POST command output (potentially containing sensitive catalog metadata, file paths, and shooting habits) to arbitrary external URLs. Additionally, the feedback command and LIGHTROOM_CLASSIC_FEEDBACK_ENDPOINT environment variable provide mechanisms for sending local data to remote servers, which is disabled by default but easily activated.
[DYNAMIC_EXECUTION]: The 'Playbook' system (described in Steps 2 and 5) executes sequences of commands stored in JSON files derived from previous sessions. These playbooks include variable substitution (slots_resolved) and represent a form of dynamic logic execution based on stored 'learnings' or session journals.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect injection surface by processing untrusted data:
Ingestion points: Reads data from external SQLite .lrcat catalog files and local data stores (data.db, teach.log, and feedback.jsonl).
Boundary markers: The instructions do not specify boundary markers or sanitization for data retrieved from the Lightroom catalog before it enters the agent context.
Capability inventory: The tool can execute shell commands via the CLI, write to files via --deliver file:<path>, and perform network operations via --deliver webhook:<url>.
Sanitization: There is no evidence of sanitization or validation of the content retrieved from the catalog before it is processed or used in the 'learning' loop to synthesize new playbooks.
[COMMAND_EXECUTION]: The skill uses bash to interact with the lightroom-classic-pp-cli binary, passing user-supplied or agent-generated arguments directly to the shell for execution.

pp-lightroom-classic