AI Engineering

Assume This Stage Is The Last One You Get advises planning each early stage of a timed research run so it leaves a valid model, score, and rerunnable script if the later stages never happen. It is used when wall-clock limits make deferred work unlikely to execute.

Auditable Method Choice helps you decide whether a more complex method fits the remaining clock by timing a small run, extrapolating full cost, and writing a go/no-go before you commit. It also covers checkpointing and split/seed control so the stronger option can fail safely.

Chemistry Ablations And Curves helps plan ablation tables and curve-based evidence when a model can’t be trained at full scale. It distinguishes component ablations from input ablations and suggests the highest-fidelity curve you can still ship.

Chemistry Group Attribution Across Splits defines how to evaluate substructure attributions over a held-out split instead of one molecule. It also compares the same attribution on a baseline model and treats learned masks as outputs.

Comparator Table Coverage helps you transcribe a source table into row-level targets, audit your inclusion filter against each row, and report one run per class the table distinguishes. Use it when study design could otherwise drop whole comparator rows.

Hunt actively tests a scoped target for vulnerabilities by reading scope and prior findings, detecting the stack, and running targeted probes with concrete payloads. Use it when you need guided recon and exploitation attempts for a specific vuln class.

Interaction Inventory Of The Modelled Complex compares reference and predicted molecular interfaces by contact class, listing recovered, missed, and spurious interactions. Use it for docked or co-folded complexes when RMSD or DockQ alone does not explain the binding chemistry.

LLM AI Hunting targets prompt injection, agent tool abuse, output-handling bugs, and model-server RCE in AI features. Use it when testing chatbots, RAG systems, coding agents, MCP servers, or other LLM-integrated surfaces.

Null Bounds First helps you report a headline estimate and a noise bound as separate results, especially after permutation or placebo tests. It also enforces value lines for every condition and keeps bounds from replacing the answer.

SAST Methodology breaks static analysis into focused agent steps for finding and verifying security bugs. Use it when you need file ranking, flow tracing, gap analysis, PoC confirmation, or adversarial validation.

Supplied Parameter Ledger turns shipped parameter files into a claim ledger, with one row per named value, range, case, or condition. Use it when a study file must be audited, reproduced, and reported in the file's own labels and units.

Hue generates design language skills from a URL, name, or screenshot so Claude Code or Codex can produce consistent brand-matched UI. Use it when creating or remixing a design system skill.