The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the superset CLI to manage hosts, agents, workspaces, and terminals. It also uses bun and bunx for development tasks within the Superset monorepo. These are standard operations for the tool's intended purpose.\n- [EXTERNAL_DOWNLOADS]: The instruction to run superset update allows for downloading updates to the CLI tool. These actions target the vendor's own services.\n- [REMOTE_CODE_EXECUTION]: The use of bunx cli-framework dev involves fetching and executing developer tooling from an external registry. This is a common practice in development environments for exercising source code against configured APIs.\n- [CREDENTIALS_UNSAFE]: The skill references .env files for configuration and authentication (superset auth whoami). This is a standard practice for managing secrets in local development environments and is not used for exfiltration to unauthorized parties.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because the coordinator agent reads output from worker agents (superset terminals read). If a worker agent's output contains malicious instructions, they could influence the coordinator.\n
Ingestion points: Terminal output is ingested via the superset terminals read command in SKILL.md.\n
Boundary markers: The skill defines specific protocol envelopes (SUPERSET_WORKER_DONE, SUPERSET_WORKER_BLOCKED) to mark the end of worker tasks, which serves as a basic boundary mechanism.\n
Capability inventory: The coordinator can execute commands, create agents, and send text to existing sessions.\n
Sanitization: No automated sanitization is described; the skill relies on the coordinator agent to 'independently verify risky or overlapping changes' and inspect snapshots for malformed envelopes.

superset-orchestrate