Search Species Agent Skill

Search Species

🔄 Core Workflow (CRITICAL)

When assisting users with chemical searches, you MUST adhere to the following step-by-step workflow. Note: Searching can be highly time-consuming; always prioritize efficiency.

• Acquire Target: Identify the chemical name, identifier, or SMILES the user wants to query.
• Select Engine: Choose the most targeted search backend (pubchem, opsin, wikidata, or all) based on the query type. Avoid using all unless strictly necessary, to minimize search times.
• Execute Search: Use the search command to query the database. You must set an appropriate max_cands limit to prevent excessively long processing times and reduce data noise.
• Evaluate Results: Carefully review the returned summary data in the output.
• Confirm & Iterate: Present the retrieved data to the user for confirmation. If the result is ambiguous or incorrect, communicate with the user to adjust the search keywords and restart the process.

______________________________________________________________________

Search Backend Overview

search-species integrates three distinct backends. Each serves a specific purpose in the chemical informatics workflow:

| Feature | OPSIN | PubChem | Wikidata |

| :------------------ | :----------------------------- | :------------------------ | :------------------------------ |

| Core Method | Algorithmic Parser | Curated Database | Knowledge Graph |

| Primary Input | IUPAC English Names | Names, CIDs, SMILES | Common & Multilingual Names |

| Molecular Image | Supported (Rendered) | Supported (Stored) | Rarely Available |

| Mass/Formula | Calculated via RDKit | Database Metadata | Database Metadata |

| Key Strength | Handles theoretical molecules. | Highly standardized data. | Vernacular & Cross-lingual. |

*(For more detailed engine capabilities, limitations, and data normalization behavior, see reference/backends.md)*

Quick Start & Command Outputs

Typical search syntax:

uvx search-species <engine> "<query>" [max_cands] -o <output_dir>

> Output: Prints the retrieved species data summary and the file path where each candidate's JSON is saved (e.g., SpeciesCandidate(...) written -> ./cache/xyz.json).

Typical render syntax:

uvx --from search-species render-species <candidate_files...> -o <output_dir>

> Output: Prints the file path of the successfully generated image card (e.g., Successfully rendered -> ./gallery/xyz.png).

Example

PubChem (Standard database lookups):

uvx search-species pubchem "benzene" 5 -o ./results

OPSIN (Theoretical molecules & strict IUPAC):

uvx search-species opsin "2-acetyloxybenzoic acid"

Wikidata (Multilingual & common/trade names):

uvx search-species wikidata "Аспирин"
uvx search-species wikidata "TNT"

Agent Checklist

When using this toolkit for users, ensure you cross-check these points with the Core Workflow:

• Engine Match: Match the engine to the query type based on the overview table.
• Data Scope: Remember this tool *only* retrieves structural identity (Name, Formula, Mass, SMILES, 2D Image).
• Fallback: If pubchem fails on a systematic name, fallback to opsin.
• Quoting: Always wrap the chemical <query> in quotes.

References

• Engine Details & Limitations: reference/backends.md