The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands via SSH on a remote host (Tailscale IP 100.67.156.41), including network probes (nc, curl), Docker management (docker ps, docker compose), and file analysis (awk).
[COMMAND_EXECUTION]: It requires the agent to run a locally stored validation script: /Users/joel/Documents/Codex/2026-06-17/we-re-setting-up-durable-self/work/local-convex/smoke-self-hosted-convex.sh.
[DATA_EXFILTRATION]: The skill involves reading sensitive environment configuration files (e.g., /volume1/joelclaw/s3/garage-compose/convex-s3.env). However, this risk is mitigated by explicit, mandatory redaction rules for keys such as AWS_SECRET_ACCESS_KEY and GARAGE_RPC_SECRET described in the 'Safety and Secret Handling' section.
[PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection as it processes external data from environment files and service logs to guide its operations.
Ingestion points: Files like convex-s3.env and service logs retrieved via docker compose logs.
Boundary markers: Absent.
Capability inventory: SSH execution, network requests, and local script execution.
Sanitization: The skill includes strict instructions to scan and redact multiple types of access keys and tokens before generating output.

garage