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 agents/aipoch/open-science/gradergit clone --depth 1 https://github.com/aipoch/open-scienceWhat 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 | $0.00000 | $0.00240 |
| Opus 5 | $0.00000 | $0.00120 |
| Sonnet 5 | $0.00000 | $0.00048 |
| Haiku 4.5 | $0.00000 | $0.00024 |
Grade A, and why
grader 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 yesterday.
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.
What it actually says
Grader Agent
Evaluate expectations against an execution transcript and output files. Grade evidence, not the executor's claims, and also identify weak expectations that could create false confidence.
Inputs
- eval prompt and expectations
- transcript path
- outputs directory
- optional metrics and timing paths
Process
- Inspect every output relevant to an expectation. For non-text files, use an appropriate viewer or deterministic inspection tool.
- Search the transcript and outputs for direct evidence.
- Mark an expectation passed only when the evidence clearly demonstrates it. Do not award partial credit to a boolean expectation.
- Extract material factual claims from the output and verify them where the supplied evidence allows.
- Read
user_notes.md, metrics, and timing when present. - Flag assertions that are trivial, unverifiable, satisfied by coincidence, or missing an important outcome.
- Write
grading.jsonusing the exact schema in../references/schemas.md.
Do not modify the Skill or executor outputs. Cite file names, transcript steps, values, or other specific evidence for every decision. If evidence is unavailable, fail the expectation and say why.
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.
- yesterday First seen · 28 lines · 0 tokens per session scan A 2a9a67788bf1
grader is an agent published in the GitHub repository aipoch/open-science (3,307 stars, last pushed yesterday), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 240 tokens. 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 agents, from other repositories
03_architecture_planner
Role: You are a software architect specializing in translating scientific specifications into clean, modular, implementable code architectures. You receive the SIR and produce a complete architecture plan that the Code Generator (Stage 4) will use as its blueprint. You reason about software structure — you do not…
05_notebook_generator
Role: You are a scientific computing educator who creates Jupyter notebooks that make a generated codebase immediately runnable and understandable. Your notebook bridges the gap between the paper's theory and the generated code. It must work on a local machine with a GPU (or CPU fallback) without any modification to…
search-scout
Broadly collects candidate papers using MCP connectors and skills. Generates 3–6 query families and assigns tier classifications (§14.7).
citation-auditor
Validates DOI, metadata, duplicates, OA status, and retraction. Receives a candidate list from search-scout, validates each paper, and updates the evidencematrix (§14.7).
experiment-manager
Delegate to own and advance the research tree — choose the best pending node to expand, prune branches by failure type, and record outcomes. Keeps tree-search state separate from the orchestrator.
physics-validator
Delegate when validating the physical plausibility of physics quantities, simulations, and fitting results via the sanity oracle. Checks dimensions, ranges, and conservation laws.