The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data through various commands, creating a surface for indirect prompt injection if ingested files contain malicious instructions.
Ingestion points: The commands shield scan, shield redact, shield ingest, and shield ask in SKILL.md accept external file paths or raw text as input for processing.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat ingested content as untrusted.
Capability inventory: The skill executes the mixlayer-pp-cli binary, which has network capabilities and can write to the local file system.
Sanitization: While the tool provides PII redaction, it does not provide specific sanitization against instruction injection within the data.
[DATA_EXFILTRATION]: The CLI includes features that allow data to be sent to external network endpoints.
The --deliver webhook:<url> flag enables POSTing command results to arbitrary external URLs.
The feedback --send command allows logs and feedback to be transmitted to an external endpoint if configured.
[COMMAND_EXECUTION]: The skill operates by executing the mixlayer-pp-cli binary through the shell.
[EXTERNAL_DOWNLOADS]: The skill guides the installation of software from external vendor repositories.
Fetches @mvanhorn/printing-press-library from NPM.
Downloads the mixlayer-pp-cli source from GitHub.
[METADATA_POISONING]: The author field in SKILL.md (Cathryn Lavery) does not match the provided author context (mvanhorn), though all linked resources belong to the vendor infrastructure.

pp-mixlayer