Spawner

Spawn sub-agents for independent work that needs file edits, builds, or long-running tasks. Use when work is too heavy for inline execution. NOT for quick re...

Security Scan

The skill's instructions, requirements, and scope are internally consistent for a 'spawner' that coordinates sub-agents and it does not request unrelated credentials, installs, or system access.

Purpose & Capability

Name/description (spawn sub-agents for long-running or file-editing tasks) matches the SKILL.md: all guidance is about role selection, exit criteria, files to provide, model routing, and runtime rules. There are no unrelated environment variables, binaries, or config paths requested.

Instruction Scope

Instructions stay within the stated purpose: they define templates, which files to inject, exit criteria, and rules for polling/retries. The skill does not instruct the agent to read arbitrary system files, exfiltrate data, or contact external endpoints. It explicitly restricts what to load from sub-agents (only output files).

Install Mechanism

No install spec and no code files — instruction-only. This minimizes disk/write risk; nothing is downloaded or installed by the skill.

Credentials

The skill requests no environment variables, credentials, or config paths. That is proportionate to its claimed function of orchestrating sub-agents.

Persistence & Privilege

The skill is not set always:true and does not request persistent system privileges. Autonomous invocation is allowed by default on the platform but the skill itself does not demand elevated persistence or config changes.

Assessment

This skill is coherent with its stated purpose and does not request credentials or install code, so it's low-risk in the narrow sense of composition. Before installing: (1) confirm how your platform implements 'spawning' — ensure spawned sub-agents are sandboxed and have limited file access; (2) restrict which models and files the agent can use to avoid unintended edits or long-running compute; (3) require explicit user consent before any file writes or network interactions performed by sub-agents; and (4) test in an isolated project/repo to verify the platform enforces the SKILL.md rules (for example, that transcripts are not automatically injected back into the parent agent). If you cannot verify those platform behaviors, exercise caution despite the skill itself appearing coherent.

Before Spawning

1. Role: Pick ONE — coder, writer, browser-automator, researcher, reviewer.
2. Exit criteria: What artifact proves done? A file path, URL, test result.
3. Context: Only inject files the agent needs. Task description under 2K tokens.

Model Routing

All sub-agents use free models (groq-llama → cerebras-llama → nim-llama → or-scout). Never spawn on premium models unless explicitly requested.

Task Template

You are a [ROLE]. Your job: [ONE SENTENCE].
Files you need: [LIST PATHS]
Exit criteria: [ARTIFACT]
If stuck after 2 attempts, STOP and report what failed.

After Completion

• Verify output file exists. If missing, retry inline or on fallback.
• Read only output files — never load sub-agent transcripts into context.

Rules

• 1 sub-agent at a time.
• Don't poll in a loop — they auto-announce.
• Don't retry on the same model that just 429'd.