OlympHill
D

Diff Addons (Grade A)Security-tested development skill for Claude AI. Grade A. Verify profile isolation by diffing the declared inventory (script) against what's actually loaded in this Claude session (system-reminder)

(0)
Daniel NikulshynReviewed by Daniel Nikulshyn·Updated July 2026

Overview

Diff Addons is a security-tested development skill for Claude AI that verifies profile isolation by comparing the declared inventory of skills, commands, and agents against what's actually loaded in the Claude session. It operates in three phases: gathering the declared inventory from the filesystem, enumerating the actual inventory from the session context, and producing a diff report. This skill helps ensure that the profile's declared components match what's actually loaded, providing a check on profile isolation.

Key features

  • Profile isolation validation
  • Three-phase process for accuracy
  • Diff report generation
  • Extraction self-checks

Pricing

Model
Free
Category
Skills
Rating
No reviews yet

Use cases

Profile Isolation Verification

Use Diff Addons to verify that a Claude AI profile's declared skills, commands, and agents match what's actually loaded in the session.

Development and Testing

Utilize Diff Addons in development and testing phases to ensure that profiles are correctly configured and isolated.

Pros & Cons

Pros

  • Validates profile isolation
  • Ensures declared inventory matches actual loaded components
  • Operates in a structured three-phase process

Cons

  • Requires manual execution and interpretation of results
  • Dependent on accurate extraction from system-reminder text
  • May require re-running due to extraction discrepancies

Reviews

Sign in to leave a review.

No reviews yet. Be the first!

Q&A

What does it not do?

It runs on macOS only at launch; Windows is planned and Linux is likely to follow. Although the Electron shell is cross‑platform, terminal launch and Keychain sync need ports. Profiles live on your machine with no cloud sync, though you can use exportProfile/importProfile to move them. No GitHub token is required; ClaudeWorks tries the gh CLI first, then GITHUBTOKEN, then falls back to anonymous fetches (60 requests per hour). Teams require Anthropic's experimental flag (CLAUDECODEEXPERIMENTALAGENT_TEAMS=1) and may change as the upstream matures. The app does not generate or manage Claude API keys—you must run claude login first. The app copies credentials from your existing login.

Asked by Quentin Lefevre · Nov 23, 2025

What does it do?

Each profile runs in its own CLAUDECONFIGDIR, which is Claude Code's built-in way to redirect its config root. MCP servers, skills, agents, and slash commands from profile A can't appear in profile B's session, providing real per-process isolation rather than config swapping. A curated plugin marketplace ships inside the app, letting you browse 114 upstream marketplaces and 647 featured plugins without leaving ClaudeWorks. One search bar queries 7,009 entries (marketplaces, plugins, skills, commands, agents, and MCP servers) at once. Profiles can include arbitrary shell aliases that change directories and optionally auto‑run /workflow or a saved prompt. One profile can claim the bare claude command, so plain claude in any terminal uses your default profile without the bloat of others. Additional per‑profile /workflow commands and named variants become real Claude Code commands in that profile's session, with directory‑scoped variants. Experimental multi‑agent Teams let you compose a team from existing profiles, assign roles, pick a lead, and launch. Per‑directory MCP management lets you disable individual servers inside sensitive repos while keeping them available elsewhere. A status bar builder offers 17 widgets you can drag to reorder, save, load, and override per profile. Profiles can be exported/imported as a self‑contained JSON.

Asked by Faisal Rahman · Nov 7, 2025

Why does this exist?

I kept managing two Claude Code setups by hand, the work one with its MCP servers and skill set, and the personal one with a different CLAUDE.md and different tools. Swapping meant editing ~/.claude.json and mcp.json by hand, then undoing it next session. Claude Code has no native multi-profile support, so I built this instead. Plugin caches are shared across profiles via symlinks, so you install a plugin once and updates propagate everywhere.

Asked by Camille Laurent · Sep 22, 2025

Ask a question

Skills alternatives

Manage Headers (Grade A) logo

Manage Headers (Grade A)

Skills

Security-tested development skill for Claude AI. Grade A. Inspects and configures the security headers a Power Pages site sends to browsers — Content Security Policy, frame and clickjacking protection

(0)
Free
Using Git Worktrees (Grade A) logo

Using Git Worktrees (Grade A)

Skills

Security-tested data-ai skill for Claude AI. Grade A. Use when starting feature work that needs isolation from current workspace or before executing implementation plans - creates isolated git worktre

(0)
Free
Ga4 Bigquery Schema (Grade A) logo

Ga4 Bigquery Schema (Grade A)

Skills

Security-tested data-ai skill for Claude AI. Grade A. GA4 BigQuery Export Schema Reference — complete field reference, nested structures, query patterns, and performance tips

(0)
Free
Meta Capi (Grade A) logo

Meta Capi (Grade A)

Skills

Security-tested data-ai skill for Claude AI. Grade A. Meta Conversions API (CAPI) Setup Reference — architecture, event types, customer information hashing, deduplication, implementation examples, AEM

(0)
Free
Callees (Grade A) logo

Callees (Grade A)

Skills

Security-tested development skill for Claude AI. Grade A. Lista o que uma funcao/metodo chama (call graph direto)

(0)
Free
Test Module Name (Grade A) logo

Test Module Name (Grade A)

Skills

Security-tested data-ai skill for Claude AI. Grade A. Name Haskell test modules after the module under test with a Spec suffix in the same namespace. Use when writing or reviewing Haskell test module

(0)
Free
Board Of Directors (Grade A) logo

Board Of Directors (Grade A)

Skills

Security-tested data-ai skill for Claude AI. Grade A. Simulate a 5-member expert board deliberation for major decisions. Use when evaluating plans, architecture choices, feature designs, or any decisi

(0)
Free
Advpl Mvc Avancado (Grade A) logo

Advpl Mvc Avancado (Grade A)

Skills

Security-tested development skill for Claude AI. Grade A. MVC avançado via PE (Pontos de Entrada) — adicionar grids customizadas em telas MVC padrão (CNTA300/MATA070/MATA440/MATA460/FINA040 via *STRU)

(0)
Free