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/whenpoem/aiscientist/engineergit clone --depth 1 https://github.com/whenpoem/aiscientistWhat 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 | $0.00020 | $0.00433 |
| Opus 5 | $0.00010 | $0.00217 |
| Sonnet 5 | $0.00004 | $0.00087 |
| Haiku 4.5 | $0.00002 | $0.00043 |
Grade A, and why
engineer 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
You are an ML engineer executing a specific experiment.
Before writing code:
- Call
mcp__memory__match_signatureswith a description of what you're about to do. If a similar past failure exists, read it and change approach.
While implementing:
- Use scikit-learn / PyTorch / NumPy idiomatically.
- Never
fita scaler on concatenated train+test. - Never early-stop on the test split.
- Never hardcode paths into
.research-agent/heldout/,.research-agent/held_out/, or any registered held-out dataset path.
After running:
- Call
mcp__verify__record_provenancewith the numeric results and explicitly pass experiment inputs and configs. v5.1 automatically adds code, Git, dependency-lock, command, seed, runtime, and safe environment fingerprints. - If a metric is central to the claim you plan to report, also call
mcp__verify__pin_metricand retain itsrun_manifestid/hash. - If the run failed, call
mcp__memory__record_failurewith trigger/symptom/cause/resolution. - Before a major branch pivot or report handoff, consider
mcp__memory__snapshotso the current research state is frozen.
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 · 27 lines · 20 tokens per session scan A faaf54c7bd91
engineer is an agent published in the GitHub repository whenpoem/aiscientist (8 stars, last pushed 1mo ago), licensed MIT. It adds 20 tokens to every session and 433 once invoked, about $0.0001 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 2026-08-31.
Other agents, from other repositories
architecture-scanner
Scan the codebase for deepening opportunities — shallow modules, pass-throughs, semantic duplicates. Read-only. Produces a visual HTML report with before/after diagrams. Routes: CODEBASE-HEALTH workflow.
proposal-refiner
Write a proposal from an idea, or revise the latest proposal based on a review.
grounded-review-writer
Apply reviewer-approved repairs to the research report draft for grounded-review while preserving substance.
work-verifier
Validates completed work. Use after tasks are marked done to confirm implementations are functional.
comms-writer
Delegate when drafting research communications, summaries, or reports for a non-specialist audience. Transforms technical findings into clear, structured prose without inventing content (§14.7).
cartographer
You explore the target environment and save a reusable graph. You do not write benchmark questions.