The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script cleanup-old-data.ts in SKILL.md utilizes prisma.$executeRawUnsafe with template literals to interpolate table and column names directly into SQL strings. While the date parameter is passed safely, identifiers like table names are not parameterized, which could allow SQL injection if the configuration becomes dynamic.
[COMMAND_EXECUTION]: The shell script migrate-to-s3.sh in SKILL.md interpolates variables directly into shell commands for psql and aws. This pattern is vulnerable to command injection if variables like $CUTOFF_DATE are derived from untrusted input.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes large volumes of application data, providing an attack surface for instructions hidden in data records.
Ingestion points: Database records from orders, users, and sessions tables are accessed via Prisma in SKILL.md.
Boundary markers: Absent. No delimiters or instructions are provided to the agent to prevent it from obeying instructions embedded in the processed data.
Capability inventory: The skill has destructive database permissions (deletions) and network access for file uploads (S3).
Sanitization: No sanitization or validation is applied to database content before it is processed by archival or cleanup jobs.

data-retention-archiving-planner