The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains hardcoded credentials for a reference MinIO instance used during setup smoke testing (minioadmin / minioadminPassWD9001).
[DATA_EXFILTRATION]: The skill provides instructions for the agent to access and read sensitive configuration files that store API keys and service secrets, including /volume1/joelclaw/s3/garage-compose/convex-s3.env and local .env files. While the skill instructs the agent not to print these in chat, the capability to read them constitutes a data exposure risk.
[COMMAND_EXECUTION]: The skill enables extensive shell command execution, including running local scripts (prepare-env.sh, run.sh), executing remote scripts on a NAS via SSH, and performing administrative tasks using sudo for Postgres and Docker management.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes local configuration files and command outputs without explicit boundary markers or sanitization.
Ingestion points: Local environment files, project notes (.svx), and outputs from network health checks (curl, nc).
Boundary markers: Absent; there are no specific delimiters or instructions to ignore embedded content within the ingested files.
Capability inventory: Local and remote (SSH) shell command execution, Postgres database administration, and file system access.
Sanitization: Absent; no validation or filtering is applied to the content read from external configuration files.

local-convex