"Routes TensorFlow Privacy users who want privacy budgets,
Scanned 9/9/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill privacy-accounting --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Privacy Accounting?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-privacy-accounting)More formats (shields.io, HTML) on the badges page.
---
name: privacy-accounting
description: "Routes TensorFlow Privacy users who want privacy budgets,
epsilon/delta statements, or the repo's privacy accounting CLIs."
disable-model-invocation: true
metadata:
disco-role: operating
license: Apache 2.0
---
# Privacy accounting
Use this sub-skill when the user wants to compute privacy budgets, find a noise multiplier, or explain the accounting assumptions behind a DP-SGD run.
## Trigger phrases
- "compute epsilon"
- "find the noise multiplier"
- "DP-SGD privacy statement"
- "privacy budget"
- "accountant type"
- "user-level privacy"
- "tree aggregation accounting"
## What this sub-skill covers
- `compute_dp_sgd_privacy` / `compute_dp_sgd_privacy_statement`
- `compute_noise_from_budget`
- the `RDP` and `PLD` accountant choices
- tree-aggregation privacy accounting helpers
- user-level privacy notes when `max_examples_per_user` matters
## What it does not cover
- training-loop integration -> `../training/`
- `DPQuery` internals -> `../queries/`
- attack analysis -> `../privacy-tests/`
- fast clipping internals -> `../fast-clipping/`
## Read this before you act
- `references/api-reference.md` for the verified function signatures and the CLI flag map.
- `references/troubleshooting.md` for missing-flag, delta, user-level, and accountant-selection failures.
- `../../references/install-and-scope.md` for the minimum runtime.
## Typical workflow
1. Determine whether the user needs an example-level or user-level statement.
2. Choose `RDP` unless the user explicitly wants `PLD`.
3. Use the bundled CLI helper when a command-line answer is enough.
4. Use the API reference when the user wants to embed the calculation in code.
5. If the user wants a tree-aggregation variant, read the tree-accounting notes before answering.
## Bundled helpers
- `scripts/compute_dp_sgd_privacy.py` mirrors the privacy-statement CLI with the same core parameters.
- `scripts/compute_noise_from_budget.py` mirrors the inverse search CLI that solves for a noise multiplier.
## Cross-links
- `../training/` usually feeds the values that this sub-skill consumes.
- `../queries/` owns the lower-level query math that eventually feeds some accounting paths.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!