Live survey304 models · 14 providers · 23,592 auditsAudited implement · SAFE →What changedHow we verifyllms.txt

Agent skill · Audited from source

literature-search-openalex

skill · google-deepmind/science-skillsApache-2.0 licenceAudited 13 Nov 2026 · source 4e5b755b9ea4

★ 3,467pushed 2 days ago
✕BLOCKGrade D · 69/100 · 8 findings · 4e5b755
3,467repo stars · whole google-deepmind/science-skills

Is it safe? No — BLOCK, grade D (69/100). We read this skill's source at commit 4e5b755b9ea4 and observed filesystem, network and shell access and no hard-coded secrets. The audit raised 8 findings, 6 critical or high. What it does: Retrieve and analyze AlphaFold predicted structures for a protein. Use when the user provides a specific UniProt Accession ID and wants structural confidence metrics (pLDDT), domain boundary analysis, or disorder…

Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
More
Cite

Cite this page

ModelsAtlas. “literature-search-openalex (google-deepmind/science-skills)”. Retrieved 13 Nov 2026. https://modelsatlas.com/skills/google-deepmind/literature-search-openalex

Blocked. The page stays public so you can check this skill if someone recommended it to you. The audit found 6 critical or high findings, starting with “Code injection” in skills/predictingthepast/scripts/run_inference.py:140. Do not install it without reading them.

3 skills are called “literature-search-openalex”. You are viewing google-deepmind/science-skills. 3 of them are BLOCK.

Trust audit

Audit 0.4.1 · source read 13 Nov 2026

Trust score69/100✕BLOCK D
0 critical6 high2 medium0 low0 info

BLOCK means the audit found something that could harm you or your machine. Grade D is shared by 749 of 18,389 audited skills.

What the source does

  • FilesystemDeclared5 observations
  • NetworkDeclared5 observations
  • ShellDeclared1 observation
  • DependenciesPinnedNothing floats
  • SecretsNone foundNo keys in source

Five audit layers

LayerWhat we checkResultFindings
L0 ProvenanceRepo owner, licence, archive status, activityPASS0
L1 Static sourceSecrets, path traversal, supply chain, hidden codeFAIL2
L2 InstructionsPrompt injection, authority framing, hidden directivesFAIL6
L3 InventoryThe files and declarations match the sourcePASS0
L4 SandboxRun in isolation and watch behaviourSKIPPED why0

Gates applied: no behavioural pass.

Findings (8)

HIGHCode injectioncode.deserialize · skills/predictingthepast/scripts/run_inference.py:140
F-02 Code injectionCWE-78 · CWE-94 · CWE-95Layer L1
checkpoint = pickle.load(f)
Why
deserialises untrusted bytes into live objects
Fix
use json or yaml.safe_load
HIGHPrompt injectionprompt.credential_read · skills/alphagenome_single_variant_analysis/SKILL.md:171
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
# Load environment variables from ~/.env
Why
asks the agent to read credentials
Fix
—
HIGHPrompt injectionprompt.credential_read · skills/alphagenome_variant_impact_score/SKILL.md:19
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
> tools**. Access to outputs requires an AlphaGenome API key subject to terms of
Why
asks the agent to read credentials
Fix
—
HIGHPrompt injectionprompt.credential_read · skills/credentials/SKILL.md:57
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
**NEVER** run `cat ~/.env`, `grep "VAR" ~/.env` (without `-q`), `echo $VAR`, or
Why
asks the agent to read credentials
Fix
—
HIGHPrompt injectionprompt.credential_read · skills/credentials/SKILL.md:85
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
printf "Enter CREDENTIAL_NAME (typing hidden): " && read -s val && echo && echo "CREDENTIAL_NAME=$val" >> "ENV_FILE" && echo "Saved."
Why
asks the agent to read credentials
Fix
—
HIGHPrompt injectionprompt.credential_read · skills/credentials/SKILL.md:90
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
All helper scripts inside the calling skills load these credentials
Why
asks the agent to read credentials
Fix
—
MEDIUMNetwork egressnet.raw_ip · skills/predictingthepast/SKILL.md:209
F-04 Network egressCWE-200 · CWE-319Layer L1
`http://127.0.0.1:<port>/<filename>`) as the primary clickable link in your
Why
No further detail recorded for this rule.
Fix
—
MEDIUMSupply chainprompt.pipe_to_shell · skills/uv/SKILL.md:39
F-08 Supply chainCWE-829 · CWE-1357Layer L2
-   **Unix/macOS**: `curl -LsSf https://astral.sh/uv/install.sh | sh`
Why
No further detail recorded for this rule.
Fix
—

Audit 0.4.1 · source sha 4e5b755b9ea4 · 13 Nov 2026Report an issue or request a re-scan

Provenance

Who published it, and what the verdict is pinned to

  • Publishergoogle-deepmind GitHub organizationOwner of the repository this skill is read from
  • Source foldergoogle-deepmind/science-skillsThe exact folder we audited
  • LicenceApache-2.0As the repository states it
  • Commit pinned for this verdict4e5b755b9ea4A new audit runs when the source changes
  • Repository statusNot archivedLast push 2 days ago
  • Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

What it tells the agent

Full source on GitHub ↗

Its description: Retrieve and analyze AlphaFold predicted structures for a protein. Use when the user provides a specific UniProt Accession ID and wants structural confidence metrics (pLDDT), domain boundary analysis, or disorder…

---
name: alphafold-database-fetch-and-analyze
description: >
  Retrieve and analyze AlphaFold predicted structures for a protein. Use when
  the user provides a specific UniProt Accession ID and wants structural
  confidence metrics (pLDDT), domain boundary analysis, or disorder
  assessment. Do not use if the user only has a protein name, gene name,
  or amino acid sequence — ask for a UniProt ID first.
metadata:
  publisher: google
---

# AlphaFold Database: Fetch and Analyze

## Prerequisites

1.  **`uv`**: Read the `uv` skill and follow its Setup instructions to ensure
    `uv` is installed and on PATH.
2.  **User Notification**: If
    .licenses/alphafold_database_fetch_and_analyze_LICENSE.txt does not already
    exist in the workspace root directory then (1) prominently notify the user
    to check the terms at https://alphafold.ebi.ac.uk/, then (2) create the file
    recording the notification text and timestamp.

## Overview

Downloads AlphaFold predicted structures (mmCIF) and Predicted Aligned Error
(PAE) matrices from the AlphaFold Database for a given UniProt ID, then performs
automated heuristic analysis on structural confidence (pLDDT), intrinsically
disordered regions, rigid domain boundaries, and inter-domain flexibility.

**Do NOT use when:**

-   The user only has a protein name, gene name, or amino acid sequence (no
    UniProt ID) — ask them to look up the ID on
    [UniProt](https://www.uniprot.org).
-   The user wants to search for structural homologs (use **Foldseek**).
-   The user wants to run AlphaFold predictions on a custom sequence.
-   The user needs experimental PDB structures (use **RCSB PDB**).

## Core Rules

-   **Use the Wrapper**: ALWAYS execute the provided helper scripts to query the
    database rather than accessing the database directly. The scripts
    automatically enforce the required rate limit gracefully.
-   Do not attempt to calculate domain boundaries or assess structural disorder
    yourself; always rely on the output provided by the script.
-   If this skill is used, ensure this is mentioned in the output.

## Utility Scripts

**1. Fetch Structure Files**

Downloads the `.cif` structure file, `_predicted_aligned_error.json`, and API
metadata JSON (`-metadata.json`) for a UniProt ID. Handles fragment fallback for
very large proteins.

Examples:

```bash
uv run scripts/fetch_structure.py P00520 -o /path/to/output/
uv run scripts/fetch_structure.py P04637 -o /path/to/custom_results/
```

Always specify `-o` with an absolute path or a path relative to the user's
project root, never a path relative to the skill directory.

**2. Analyze pLDDT Confidence**

Reads pLDDT confidence metrics from a saved AFDB metadata JSON file (produced by
`fetch_structure.py`) and prints a heuristic confidence assessment (structured,
disordered, mixed).

Example:

```bash
uv run scripts/analyze_plddt.py ./data/AF-P00520-F1-metadata.json
```

**3. Analyze PAE / Domain Boundaries**

Reads a downloaded PAE JSON file and detects rigid domain boundaries using a
sliding-window PAE heuristic.

Example:

```bash
uv run scripts/analyze_pae.py ./data/AF-P00520-F1-predicted_aligned_error_v6.json
```

## Interpreting the Output

The script prints analysis to stdout. Read it carefully and synthesize the
results for the user:

1.  **Isoform / Large Protein Warning (MANDATORY):** Check the script output for
    any `[!] WARNING` lines. If the script reports that no canonical entry was
    found and an isoform was used, or if the protein is very large (>2700 AAs),
    you **MUST** prominently relay this warning to the user. Do not omit this
    warning.
2.  **Synthesize the Structural Analysis**: Combine the "pLDDT Conclusion" and
    the "PAE Structural Conclusion" into a single, cohesive overall summary.
    Describe the protein's overall folding confidence, the presence of
    disordered regions, and its rigid domain layout.
3.  Highlight the supporting metrics:
    -   Overall Global pLDDT and the breakdown of fraction confidence
        (especially Very Low vs. Very High).
    -   Domain Boundary Analysis (number of distinct global domains and their
        specific residue ranges).
4.  **Explicit Disorder Warning:** If the analysis concludes that the protein is
    highly intrinsically disordered (e.g., high fraction of <50 pLDDT or lack of
    rigid domains), issue a separate, prominent warning. Advise the user against
    proceeding with whole-protein downstream structural analysis (like Foldseek
    or docking). If small ordered domains exist amidst the disorder, advise the
    user to restrict any future analysis strictly to those specific residue
    boundaries.
5.  Remind the user that per-residue pLDDT is embedded in the B-factor column of
    the downloaded mmCIF file.

Excerpt, first 4,785 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

Install

Pinned to the audited commit 4e5b755

npx skills add google-deepmind/science-skills/
uv run scripts/fetch_structure.py P00520 -o /path/to/output/
uv run scripts/fetch_structure.py P04637 -o /path/to/custom_results/
uv run scripts/analyze_plddt.py ./data/AF-P00520-F1-metadata.json
uv run scripts/analyze_pae.py ./data/AF-P00520-F1-predicted_aligned_error_v6.json
uv run scripts/alphagenome_atlas_links.py variant "chr9:128225994:G>A" \

Commands as the repository documents them: shown, never run by us. Our verdict covers commit 4e5b755b9ea4 only.

Works in which agents?

What its own text says, not a test

AgentStatusMore
Claude CodeMentioned in its text, not testedSkills for Claude Code · About Claude Code
CodexMentioned in its text, not testedSkills for Codex · About Codex
CursorMentioned in its text, not testedSkills for Cursor · About Cursor
Gemini CLIMentioned in its text, not testedSkills for Gemini CLI · About Gemini CLI
“Mentioned” means the skill’s text names the agent. We have not run it there.

Audit history

Every audit, every source change

  1. ✕BLOCK D 69 / 100

    first audit · source 4e5b755b9ea4

README badge

For maintainers · shows the latest audit

ModelsAtlas: BLOCK DMODELSATLASBLOCK · D
[![ModelsAtlas: BLOCK D](https://modelsatlas.com/skills/google-deepmind/literature-search-openalex/badge.svg)](https://modelsatlas.com/skills/google-deepmind/literature-search-openalex/)

The badge shows our verdict and grade for this skill, and links back to this page. It changes when the audit does.

Add this badge to your README

[![ModelsAtlas: BLOCK D](https://modelsatlas.com/skills/google-deepmind/literature-search-openalex/badge.svg)](https://modelsatlas.com/skills/google-deepmind/literature-search-openalex/)

Questions about literature-search-openalex

What does the literature-search-openalex skill do?
GDM Science Skills to speed up agentic scientific workflows with better grounding and higher token efficiency. Integrate insights from AlphaGenome, AFDB, UniProt and 30+ other databases and tools.
Is literature-search-openalex safe to install?
No — not without reading the findings first. The audit graded it D (69/100) and found 6 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What can literature-search-openalex access on my machine?
The audit observed that it reaches the network, runs shell commands and reads or writes files. Each of those is consistent with what it says it does. Secrets in the source: none found.
Which assistants does literature-search-openalex work with?
Its documentation mentions claude-code, codex, cursor and gemini-cli. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (4e5b755b9ea4), read on 2026-11-13. The repository is watched, and a new audit runs when it changes — this is the first audit.

Sources

Spotted something wrong? Report a correction · Methodology

Keep going in your AI

Opens with this page as the source. We add nothing else to the prompt.