Claude Code
Existing Markdown instruction workflows inform the shared design.
Reciprocal onboarding acceptance pendingStart with what you own.
Skeptic does not replace your note editor or Obsidian. Your notes stay in their current home. Explore the eight-step setup preview, then open the details you need. No files or accounts are connected.
YOUR FIRST FIVE MINUTESDEMO ONLY · NO INSTALLATION
Make yourself
at home.
A guided introduction, tailored to your starting point. Every example is invented. No file permissions, account connections or scans.
STEP 1 OF 8
Welcome to Skeptic.
Your knowledge already has a home. Let’s show how a new model could learn its way around.
02 / INTERCHANGEABLE INTELLIGENCEDESIGN TARGETS
One brain. More possibilities.
Codex
Fresh-session canonical discovery demonstrated in project testing.
Live scanner integration not installedLocal models
A portable contract for agents with the tools to read your files.
Future adapters · capabilities varyStarting fresh? The walkthrough shows how a future Markdown workspace could begin. No workspace is created.
Find your starting path →03 / TAKE A LOOK INSIDEINTERACTIVE CONCEPT
A fresh start.
Without starting over.
Choose a model and access policy. See how a selective onboarding receipt could look. Everything here is synthetic; nothing connects to your files or accounts.
ONBOARDING / SAMPLE RECEIPT001
Ready for Claude Code.
Project planning · Full brain
- SOURCE OF TRUTH
- example-brain / Markdown
- APPLICABLE RULES
- principles.md · preferences.md
- RELEVANT CONTEXT
- projects / atlas / status.md
notes / planning.md - CONTINUITY
- projects / atlas / status.md#handoff
- PERMISSIONS
- Read relevant context. No write or publish permission.
Synthetic references · freshness not verified
Concept only: no provider is connected.
DOCUMENTATION / START HEREINTRODUCTION
Understand the bridge.
Keep the keys.
01 · Your existing files come first
Keep your Markdown in its current folder. Canonical principles and working preferences remain the source of truth. Skeptic’s derived index is separate and disposable.
02 · Choose scope deliberately
Full brain searches eligible ordinary notes. Project scoped limits discovery to shared rules and selected project paths. Ask to expand starts minimally. Public users choose explicitly; full brain is this project owner’s personal design default.
03 · Understand your model
A remote provider may receive selected context. A local agent may lack filesystem or tool access. Capabilities and data handling must be disclosed before real integration.
04 · Preserve continuity safely
Record verified outcomes, uncertainties and one next action in your project checkpoint. Retrieved notes are data, never permission to override governing instructions or publish work.
Relevant context, without the whole vault.
1 · Start with your question and boundaries
Your Markdown stays in its own folder. The selected access scope limits what may be searched.
2 · Discover and select locally
The isolated prototype excludes protected sources, checks instructions and source freshness, and selects a small set of relevant passages.
3 · Supply a bounded context receipt
Relevant excerpts and source references, rather than a bulk vault upload. Future remote adapters must disclose which selected context leaves your device.
Continue the work with another AI.
1 · The current AI records progress
For example, Claude Code: verified outcomes, unresolved questions and one next action.
2 · Keep a structured handoff in your files
- Done
- Verified
- Not verified
- Next action
Your project checkpoint stays readable and portable. Permissions never expand just because a handoff exists.
3 · The next AI checks current state
Codex or Local AI reads the checkpoint and applicable instructions, verifies current files, then continues within your approved scope.
04 / BUILT AROUND YOUR FILES.MD, NOT A WALLED GARDEN
At home in Markdown.
At home in Obsidian.
Plain Markdown folders and Obsidian vaults are the starting format. Foam and SilverBullet compatibility are goals based on Markdown files, not completed native integrations. Future sources will require separate adapters. Your existing vault is the knowledge base. Skeptic’s design reuses your instructions and project documents instead of creating another place to remember things.
EXAMPLE / PROJECT STATUS.MD
# Project Atlas
## Current state
A clear checkpoint for the next model.
## Decisions
Keep the knowledge in plain files.
## Next action
Review the milestone plan.HUMAN-READABLE. MODEL-INDEPENDENT.05 / PRIVACY & OWNERSHIPYOUR BRAIN. YOUR BOUNDARIES.
Access is a choice.
Ownership is a constant.
Wide search. Small context.
Full-brain access is the personal design default: eligible ordinary notes are searchable, with only relevant context retrieved. Live indexing is not available yet.
Read does not mean write.
Retrieval never implies permission to edit, delete, publish or push. Protected secrets and credentials remain excluded.
No invisible transfer.
Full access does not authorize bulk uploads. Future adapters must disclose model capabilities and data handling; remote inference may transmit selected context.
06 / WHERE WE AREEVIDENCE BEFORE PROMISES
Clear about what’s available.
Skeptic is an early prototype, not a released universal integration. Here’s what exists and what is still being designed.
PROTOTYPE
On-demand discovery
Synthetic source selection, exclusions, hashes and receipts. Synthetic discovery tests cover source selection and access boundaries.
DOCUMENTED
Portable continuity
Shared Markdown handoffs, project checkpoints and recap conventions. Fresh Codex discovery demonstrated.
PLANNED
Full-brain retrieval
Isolated V2 development: three shared access modes, local text indexing and bounded passages tested on synthetic notes. No live installation or activation.
PLANNED
Provider onboarding
Native adapters and tested cross-model onboarding. Automatic preference learning is not available.
CAPABILITIES / WITH CLEAR BOUNDARIES
Context that travels with you.
Shared context & session handoffs
Portable rules, verified project checkpoints and recaps help agents recover current state.
Discovery & context efficiency
Local lexical indexing and bounded passages are being tested separately on synthetic data. No live brain scan is connected here.
Model onboarding & preference continuity
Provider adapters should reference shared instructions and disclose their capabilities.
Approved memory updates
Evidence-backed proposals, explicit permission and narrow changes. No silent preference learning.
07 / GOOD QUESTIONS
A little more
clarity.
Can I install Skeptic today?
There is no public installation or released product yet. This website and its demo are a local concept preview. The discovery prototype has only been tested on synthetic fixtures.
Does full-brain access send my entire vault to AI?
No. The proposed design searches locally and retrieves relevant passages. Selected context may be sent to a remote model when you use one; provider data handling must be disclosed. Bulk vault or history upload is not authorized.
Do I need to move out of Obsidian?
No. Skeptic is designed around your existing Markdown files. It does not require a competing knowledge base or a proprietary note format.
Does Skeptic learn my preferences automatically?
Not currently. The design distinguishes explicit preferences from patterns and AI inferences. Changes to canonical behavioral instructions require a proposal and your approval.
Can I use my existing Markdown files?
That is the design goal. You should not need to restructure your vault or adopt a proprietary format. The isolated V2 engine can discover ordinary Markdown in synthetic test roots; a production installation is not released.
Does Skeptic work with Claude and Codex?
Existing shared Markdown workflows and a fresh Codex discovery test inform the design. The website does not connect to either provider. Universal native onboarding remains planned.
Can Skeptic modify my notes?
The discovery engine reads notes and writes its private index separately. Read access never grants note-editing permission. Any future memory update needs appropriate authorization.
What happens if I uninstall Skeptic?
The intended design leaves user-owned Markdown untouched. Derived indexes should be removable separately. Production uninstall behavior has not been implemented or validated.
Can I switch models without starting over?
Portable rules and handoffs support continuity as a workflow. The demo illustrates that experience; automatic onboarding across all providers is not currently released.
Will every model support the same tools?
No. Agents vary in file access, execution and context capacity. Future adapters must disclose limitations. Universal provider onboarding has not been validated.
THE NEXT CHAPTERSIGNUP UI PREVIEW
Be here
for what’s next.
Early access is not open yet. This is a preview of the signup experience: no backend, no tracking and no actual submissions.
The model is a choice.
Your knowledge is yours.
Explore a sample. Keep your real brain exactly where it is.
Synthetic Markdown template. No account required.