pp-kie
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPERSISTENCEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to install the CLI binary using
npx -y @mvanhorn/printing-press-library. This is a deployment method targeting a package within the author's identified vendor namespace. - [PERSISTENCE]: The skill documents an 'automatic learning loop' that journals command usage and stores 'learnings' in a local SQLite database (
data.db) under the user's data directory. This system is designed to maintain context and optimization data across independent agent sessions. - [INDIRECT_PROMPT_INJECTION]: The skill implements a 'recall' mechanism that retrieves previously stored instructions and 'playbooks' based on user queries.
- Ingestion points: Data enters the system via the
teachandplaybook amendcommands described inSKILL.mdwhich process user queries and session notes. - Boundary markers: The instructions do not define explicit delimiters for separating retrieved 'learnings' from system instructions.
- Capability inventory: The skill uses the
Bashtool to execute thekie-pp-cli, which possesses capabilities for local file system management (configuration/state directories) and network communication (API calls and optional webhooks). - Sanitization: The skill instructions include a manual 'PII rule' advising the agent to strip personal identifiers before committing data to the learning store.
- [DYNAMIC_EXECUTION]: The learning loop automatically synthesizes 'playbook candidates' from session journals. These candidates contain structured command steps that the agent is directed to replay in subsequent interactions.
- [DATA_EXFILTRATION]: The CLI includes a
--deliver webhook:<url>flag, which allows users or agents to route command output to a specified HTTP endpoint. This is a documented administrative feature for workflow integration.
Audit Metadata