The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions mandate storing the neonConnString (database credentials) directly within the prompt text of cloud-scheduled tasks. Task metadata and prompt descriptions in scheduling services are often not encrypted or are visible in administrative logs, presenting a high risk of credential exposure.
[COMMAND_EXECUTION]: The skill executes multiple internal vendor scripts (e.g., ensure-deps.sh, apply-schema.mjs, playbook.mjs, db.mjs) using bash and node commands from the plugin's shared directory.
[PERSISTENCE]: The skill establishes persistent, unattended execution by creating two cloud-scheduled routines (Playbook-Jupi — catchup and Playbook-Jupi — daily) which run on cron-like cadences.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided documents and dossiers, creating a vulnerability surface for instruction injection. 1. Ingestion points: playbookSources (arbitrary user documents) and dossierSource (external data files). 2. Boundary markers: Employs a soft constraint directing the agent to treat document content as data rather than instructions. 3. Capability inventory: Includes filesystem access, database writes, and autonomous task scheduling. 4. Sanitization: No robust sanitization or validation is specified for data extracted from the source documents.

setup-playbook-jupi