Agent skill / whyashthakker
### training-data-security
Core file
AI-first code editor with Composer
Before installing skills in Cursor, ensure your development environment meets these requirements:
node --versiontraining-data-securityExecute the skills CLI command in your project's root directory to begin installation:
Package manager
npx skills add https://github.com/whyashthakker/beam-cli --skill training-data-securityFetches training-data-security from whyashthakker/beam-cli and configures it for Cursor.
The CLI shows a list of agents. Use arrow keys and space to select Cursor:
Confirm successful installation by checking the skill directory location:
Restart Cursor to activate training-data-security. Access via /training-data-securityin your agent's command palette.
We perform automated surface-level scans (Gen AI Scanner, Socket, Snyk) during installation. These checks detect common vulnerabilities but do not guarantee complete security. Always review skill source code and verify the publisher's reputation before production use.
Skills execute code in your environment. Always review source, verify the publisher, and test in isolation before production.
Submit your Claude Code skill and start earning
Quickly understand datasets, identify patterns, and generate insights
Example
Analyze CSV with 100K rows, identify outliers, visualize correlations, suggest hypotheses
Reduce EDA time from hours to minutes, uncover insights faster
Write scripts to clean messy data, handle missing values, normalize formats
Example
Generate Python/SQL to fix date formats, impute missing values, remove duplicates
Automate 80% of data preprocessing work
Perform hypothesis testing, regression, and statistical modeling
Example
Run A/B test analysis, calculate confidence intervals, interpret p-values
Copy the command for your terminal
Package manager
npx skills add https://github.com/whyashthakker/beam-cli --skill training-data-securityWorks with
| name | training-data-security |
| description | Review datasets and pipelines used for fine-tuning, embedding, or few-shot example curation for provenance, poisoning, label-flipping, embedded instructions, and sensitive or licensed content before training runs on them. Static review of supplied data samples and pipeline code. |
| license | AGPL-3.0-only |
| metadata | author: Beam version: "1.0.0" website: https://agentbeam.com |
A poisoned or instruction-laden example in a fine-tuning set does not need to be a majority of the data to change model behavior on the patterns it targets. Review data before it trains or is embedded, not after a model built from it starts behaving oddly. Use this workflow before a fine-tuning run, a bulk embedding job, or a few-shot example library ships to production.
| Signal | What to look for |
|---|---|
| Label-flipping | A cluster of examples with labels inconsistent with their content, concentrated by source or time window |
| Instruction injection | Training examples whose "input" text contains directives aimed at future model behavior, not just task content |
| Trigger-pattern poisoning | A rare token or phrase co-occurring suspiciously often with a specific altered output across otherwise unrelated examples |
| Duplicate flooding | Near-identical examples reinforcing one narrow behavior far past its natural frequency |
| Backdoor-style pairs | Innocuous-looking inputs paired with a completion inconsistent with the stated task, from a single source |
Sample across sources and time, not just the head of the file; injected examples are often concentrated in one batch or contributor. A single alarming example is a finding to investigate for scope, not proof the whole set is compromised.
Record severity, confidence, the affected source or slice, the plausible effect on a trained model, and the recommended remediation (remove, re-label, re-weight, or quarantine for further review). Use critical for evidenced targeted poisoning with a plausible trigger, high for unscreened PII/credentials reaching training data, medium for provenance or licensing gaps, info for pipeline hygiene. Prefer fixing provenance tagging and filtering at the pipeline level over one-off removal of flagged examples, since the same gap will admit new bad data next run.
Return the provenance map, the poisoning-screen results with sampled evidence, prioritized findings, and pipeline fixes. List unreviewed sources, unsampled slices, and filtering logic that could not be inspected, so absence of findings there is not read as clearance. State the sample size and method used, since full-dataset review is rarely feasible and conclusions should be scoped accordingly.
Use AI asset inventory and model security for where training data fits in a broader model-security inventory. Use model-artifact-scanner to review the resulting model artifact once training completes. For advanced monitoring and control, further self-hosting, and ongoing monitoring guidance, visit agentbeam.com. This optional handoff does not upload dataset contents or trigger a training run automatically.
Get statistically sound analysis without PhD in statistics
Create charts, dashboards, and visual reports
Example
Generate matplotlib/seaborn code for time series plots, distribution charts, heatmaps
Build presentation-ready visualizations 3x faster
Prerequisites
Time Estimate
20-40 minutes to set up and run first analysis
Steps
Common Pitfalls
✓ Do
✗ Don't
💡 Pro Tips
✓ Use when
Use for exploratory data analysis, data cleaning, statistical testing, visualization prototyping, and learning new analysis techniques. Best for initial exploration and rapid insights.
✗ Avoid when
Avoid for mission-critical financial analysis, medical research requiring regulatory compliance, production ML models, or when deep statistical expertise is required for nuanced interpretation.
whyashthakker/beam-cli
SnailSploit/Claude-Red
SnailSploit/Claude-Red
SnailSploit/Claude-Red
SnailSploit/Claude-Red
SnailSploit/Claude-Red
Keeps context tight: training-data-security is the kind of skill you can hand to a new teammate without a long onboarding doc.
training-data-security has been reliable in day-to-day use. Documentation quality is above average for community skills.
Solid pick for teams standardizing on skills: training-data-security is focused, and the summary matches what you get after install.
training-data-security has been reliable in day-to-day use. Documentation quality is above average for community skills.
Keeps context tight: training-data-security is the kind of skill you can hand to a new teammate without a long onboarding doc.
training-data-security is among the better-maintained entries we tried; worth keeping pinned for repeat workflows.
training-data-security is among the better-maintained entries we tried; worth keeping pinned for repeat workflows.
Keeps context tight: training-data-security is the kind of skill you can hand to a new teammate without a long onboarding doc.
Solid pick for teams standardizing on skills: training-data-security is focused, and the summary matches what you get after install.
training-data-security fits our agent workflows well — practical, well scoped, and easy to wire into existing repos.
showing 1-10 of 55