Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/understudylabs/understudy-agent-tools/design-simulated-environmentnpx skills add understudylabs/understudy-agent-tools --skill design-simulated-environmentgit clone --depth 1 https://github.com/understudylabs/understudy-agent-toolsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/understudylabs/understudy-agent-tools/design-simulated-environment)<a href="https://agentmods.dev/skills/understudylabs/understudy-agent-tools/design-simulated-environment"><img src="https://agentmods.dev/badge/skills/understudylabs/understudy-agent-tools/design-simulated-environment.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00090 | $0.04105 |
| Opus 5 | $0.00045 | $0.02053 |
| Sonnet 5 | $0.00018 | $0.00821 |
| Haiku 4.5 | $0.00009 | $0.00411 |
Grade A, and why
design-simulated-environment scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 277 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Design a simulated environment + validator
The coding agent writes this fresh for each workload — every workload's tools, state, and success criteria differ, so this skill is the recipe, not a fixed implementation. The goal: a deterministic, synthetic environment where any candidate model can run the whole agentic task and be scored on the final state, so you can compare frontier vs. local and hill-climb the local model.
Default environment first
Before building a codebase-specific environment, every new local Understudy should run in a tiny default environment so the user has an immediate, repeatable baseline:
- Task shape: read a small synthetic inbox / ticket queue / project board, decide what matters, write one structured update, and avoid forbidden writes.
- Tools:
list_records,read_record,write_note,update_status, andfinish. All tools mutate in-memory JSON only. - Gold state: the exact notes/statuses a correct run should produce, plus forbidden writes that must not occur.
- Validator axes: required-write recall, unnecessary-write precision, policy compliance, schema validity, recoverable errors, step count, latency, and cost.
- Oracle: a scripted correct trajectory must score 1.0 before any model is compared.
This default env is not the customer's app and must not be presented as proof that local can replace the incumbent. It is the first-run calibration surface: "my local model can act in a deterministic tool world, and I can compare it to a frontier model." After that, inspect the repo/traces and build the workload-specific environment below.
Why simulate (the lesson that forces this)
A recorded replay (serve the teacher's captured tool_results back) only works for a model that reproduces the teacher's exact tool path. A different or smaller model takes its own reasonable trajectory and immediately "diverges" — there is no recording for the tools it actually called. So recorded replay can't fairly test a different brain. A simulated environment implements the tools against seeded state, so every call returns a real (synthetic) result and the run is judged by what got written, not by matching the teacher.
What ships with it
16 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- examples/event-categorizer/configs/endpoints.toml 803 B
- examples/event-categorizer/convert_captures.py 9.8 KB runs code
- examples/event-categorizer/demo_gate.py 3.2 KB runs code
- examples/event-categorizer/event_categorizer.py 7.4 KB runs code
- examples/event-categorizer/mock_model_server.py 5.5 KB runs code
- examples/event-categorizer/playbook-variant.md 1.3 KB
- examples/event-categorizer/playbook.md 1.3 KB
- examples/event-categorizer/pyproject.toml 560 B
- examples/event-categorizer/README.md 4.4 KB
- examples/event-categorizer/smoke.py 4.8 KB runs code
- examples/event-categorizer/tasks.jsonl 4.3 KB
- reference.md 6.2 KB
- references/automationbench-adaptation.md 1.3 KB
- references/cookbook-traces-to-env.md 8.5 KB
- references/resumable-environment-goal.md 1.7 KB
- references/trace-to-benchmark-foundry.md 3.0 KB
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 277 lines · 90 tokens per session scan A dc6bd24279de
design-simulated-environment is a skill published in the GitHub repository understudylabs/understudy-agent-tools (16 stars, last pushed 2d ago), licensed MIT. It adds 90 tokens to every session and 4,105 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
mem0-oss-to-platform
Plan and then execute a migration of a project from the mem0 open-source / self-hosted SDK (the local Memory class) to the mem0 Platform / hosted / managed SDK (the MemoryClient class). Use this whenever a developer wants to move, switch, or migrate their mem0 usage off OSS/self-hosted to the hosted API — e.g.…
Cortex
Operate Cortex, the LifeOS memory system — the typed Knowledge Archive (People, Companies, Ideas, Research with typed related: links) plus recall of prior work sessions, ISAs, and conversations. Search, add, harvest, develop, ingest, distill, graph-navigate, recall. USE WHEN cortex, knowledge, knowledge base, search…
auditing-subgroup-fairness
Audit an OpenMed NER or de-identification model for performance disparities across demographic subgroups (sex, age band, race/ethnicity when available) using openmed.eval.fairnessreport. Use when the user wants per-subgroup recall and leakage, wants to check whether de-identification under-protects a group, wants to…
agent-memory
../../../engineering/agent-memory/skills/agent-memory/SKILL.md.
memory
Use when the user asks to remember, recall, forget, update, search, or inspect durable OpenSquilla memory, including profile facts in USER.md and long-term notes in MEMORY.md or memory//.md.
ha-data-stores
Map of Hope Agent's local data stores and safe read-only query workflow. Use when the user asks where Hope Agent stores data, wants to inspect sessions/messages/memory/logs/background jobs/knowledge indexes/settings, asks the model to query local app data, or debugging requires checking persisted state. Trigger…