Use when creating or revising a custom agent type, when an experiment needs an agent class that does not yet exist in the workspace, or when the agent design must be sized against a simulation budget.
SkillScore75Maintained
Automatic
◆AI Engineer pack
Last verified 9/24/2026
What it changes
Passed 15 of 20 test cases with the skill, and 5 of 20 without it.
Without skill25%
With skill75%
Checks it passed that the model alone missed
Recommends inheriting from PersonAgent
Requires declaring to_workspace, ask, and step as async def
Recommends saving persistent agent state inside to_workspace
and 14 more in the full report
12 fixed6 passed either way6 still failing0 broke
Pass-rate lift
+50 pp
Tokens
+69%
Agent turns
0%
Response time
-23%
Model gemini-3.6-flashMeasured 9/24/2026v218 of 24 cases passed with skill · 20 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 12 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 12
Prompt
What is the exact CLI command and options to run AST and async structure checks on a freshly written agent file located at custom/agents/finance/trader.py? Provide the command line invocation.
Mentions the --json flag for machine-readable validator output
Missed
Passed
case-19 · gemini-3.6-flash · 9/24/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md196 lines7 filesincludes scripts
Create Agent
Add Python modules under custom/agents/ relative to the workspace root (the directory that contains custom/). The backend scanner loads every *.py under that tree except paths under an examples/ segment.
Nested folders are allowed (e.g. custom/agents/research/lab_agent.py).
Prefer one primary agent class per file for clarity; the scanner can surface multiple AgentBase subclasses in the same file if you define them.
When to Use
User asks to create a new agent type (e.g., "student agent", "donor agent")
Designing agent behaviors or implementing AgentBase/PersonAgent subclasses
experiment-config needs an agent class that does not yet exist in the workspace
Extending an existing agent with custom logic
The simulation size, step budget, or runtime budget needs to shape the agent design
Do NOT use when:
You only need to discover existing agents (use scan-modules)
The task is about environment modules, not agents
Quick Reference
Use the Python interpreter from .env. See CLAUDE.md for setup.
Scale Budget
Collect the simulation scale budget before locking the agent design:
target agent count or range
expected step budget
runtime or compute budget
preferred complexity tier, such as lean, balanced, or rich
If the budget is still open, ask a single round of clarifying questions. Present 2-3 approaches with trade-offs and a recommendation, then choose the option that keeps per-agent logic proportional to the simulation size.