synthetic-sciences/openscience is an AI workbench that carries out scientific research by reading papers, forming hypotheses, writing and running code, conducting experiments, analyzing results, and preparing reports. Researchers use it for work in machine learning, biology, physics, and chemistry with remote or local models. Catalogue add-ons extend its scientific workflows through skills and instructions.
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 skills add synthetic-sciences/openscience --skill delegationgit clone --depth 1 https://github.com/synthetic-sciences/openscienceWrote 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/synthetic-sciences/openscience/delegation)<a href="https://agentmods.dev/skills/synthetic-sciences/openscience/delegation"><img src="https://agentmods.dev/badge/skills/synthetic-sciences/openscience/delegation/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/synthetic-sciences/openscience/delegation"><img src="https://agentmods.dev/badge/skills/synthetic-sciences/openscience/delegation.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00118 | $0.01612 |
| Opus 5.5 | $0.00047 | $0.00645 |
| Sonnet 5.5 | $0.00024 | $0.00322 |
| Haiku 4.5 | $0.00012 | $0.00161 |
Grade A, and why
delegation 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 19d 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 — 123 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Delegation
A worker is a second agent that starts with an empty transcript. It can do independent work in parallel and bring back a decision-ready handoff; it cannot read your mind, your conversation, or the user's intent. Delegation pays when the task has a clean boundary and a brief can carry everything the worker needs. Otherwise it costs latency, tokens and coherence, and the work is better done here.
When to delegate
Delegate when all of these hold:
- The work is independent of the step you are on: surveying an unfamiliar codebase, running and reporting an isolated experiment, drafting a section from finished results, auditing a long artifact, reproducing a claim in parallel with other work.
- You can write down the goal, the inputs, the constraints and the definition of done without reference to "what we discussed".
- The result can be checked when it comes back (a number to compare, a file to open, a report against a checklist).
- The Delegation setting allows it (Off means do it here; Auto means when it clearly helps; High means prefer it for parallelizable phases).
Never delegate:
- the literature retrieval loop: ranking candidates is the judgement, and a worker ranks for a different question (load literature-review and run it yourself);
- a step inside an experiment loop already underway, where the next action depends on the last result you saw;
- anything requiring the user's answer to a question the worker cannot ask.
Choosing the worker
| Need | subagent_type |
|---|---|
| Read and report: codebase survey, artifact audit, options analysis, literature scouting | explore (read-only) |
| Independent critical read of a draft or analysis | explore, briefed with the peer-review skill and the angle to take |
| Build or run: pipelines, data processing, scripts, files, results | data |
| Domain-heavy phase: training or fine-tuning setup, omics pipeline, PDE solver, docking | ml, biology, physics, chemistry |
subagent_type is an agent name; the Task tool's description lists the ones configured in
this installation. There is no separate specialist parameter and no execute or
critique profile. A specialist keeps the Research contract and gains its domain skill
index and domain tools (the biology specialist can query UniProt, PDB, Ensembl, KEGG,
PubMed directly). Use one when most of the phase sits in that domain; for mixed work the
data worker, briefed to load a skill, is enough. Pass background: true for work you do
not need before your next step; its result wakes you when it lands.
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.
- 19d ago Changed · +22 lines · +17 tokens per session fb23b22b6fc5
- 1mo ago Changed · +3 lines 44ec20b7a98e
- 1mo ago First seen · 98 lines · 101 tokens per session scan A f8e7a00d2ead
delegation is a skill published in the GitHub repository synthetic-sciences/openscience (4,004 stars, last pushed 3d ago), licensed Apache-2.0. It adds 118 tokens to every session and 1,612 once invoked, about $0.0005 per session on Opus 5.5. 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 2026-09-14.
Other skills, from other repositories
google-calendar
Google Calendar via gws: list events, create, accept, find free time.
google-drive
Google Drive via gws: search, list, upload, download, share.
google-forms
Google Forms via gws: create forms, add items, read responses.
google-meet
Google Meet via gws: create spaces, fetch join links, list recordings.
phone-call
Place outbound AI phone calls via Bland AI: book reservations, make appointments, ask businesses questions, then report back the transcript and summary.
people-crm
Turn LinkedIn connections (or any contact export) into a queryable personal CRM using the agent database.