-
Notifications
You must be signed in to change notification settings - Fork 0
Home
OpenSocrates is a local, open-source reasoning framework for Claude, Codex, OpenCode, Grok Build, Cursor and Google Antigravity. It provides 48 authored reasoning methods with complete English and Korean procedures. Agents retrieve eligible methods when a judgment changes; factual or mechanical work can pass through unchanged.
This wiki is the conceptual and operational guide. The repository README, SECURITY.md, CONTRIBUTING.md, tagged releases, and versioned source remain authoritative. If the wiki and a tagged release disagree, follow the tagged release documentation.
- Install or upgrade: README
- Understand the system: Architecture
- Check host behavior and limitations: Host-Support-and-Installation
- Contribute safely: Development-and-Release-Workflow
- Interpret validation claims: Validation-and-Evidence
- See active direction: Roadmap
- Report a vulnerability: SECURITY.md
- Supplies discovery guidance through the host's supported entry point.
- Retrieves eligible complete methods when the active agent encounters a judgment.
- Reconsiders the selection when the judgment changes within the same request.
- Preserves the user's goals, permissions, evidence requirements and completion conditions.
- Keeps selection, delivery, agent-reported reading and actual application distinct.
The v1.3.1 one-shot native command accepts a complete multi-line JSON document;
--stream accepts one JSON object per line. The default decision route adds no
separate selector model call. The older selector/artifact adapter remains available
for compatibility and is described separately in Architecture.
The integration fails open: a timeout, unavailable selector, invalid output, unsafe context, or non-intervention decision does not block the user's task.
- Selective, not universal: intervene only when structured reasoning is useful.
- Complete procedures: use authored theory, steps, and examples rather than method-name hints.
-
Evidence boundaries: the retained legacy Claude adapter can verify a read callback; the current decision route keeps
applied: unverifiedand does not certify semantic application. - Local integration: OpenSocrates has no backend, adds no telemetry, and uses the user's existing host login.
- Bounded claims: implementation, local validation, release validation, and live host receipts are reported as different evidence levels.
- Privacy by construction: OpenSocrates records do not retain raw prompts, transcripts, credentials, workspace content, or selector reasoning.
- Fail-open availability: OpenSocrates does not execute the user's requested task and does not prevent the host from continuing when selection is unavailable.
| Path | Purpose |
|---|---|
src/opensocrates/ |
Shared Python runtime and selectors |
content/ |
Policies, localized messages, and 48 authored methods |
plugin-src/claude/ |
Claude plugin templates |
plugin-src/codex/ |
Codex plugin templates |
schemas/source/ |
Canonical schema definitions |
schemas/v1/ |
Generated public schemas |
installer/ |
Dependency-free Node.js installer |
tools/ |
Build, validation, lifecycle, security, and release tooling |
docs/ |
Reproducible validation procedures and evidence receipts |
The current patch is v1.3.1. Read the release record and official homepage. Historical model evaluations retain their failures and source identities; this patch makes no general quality, speed or cost improvement claim.
- Use a bug report for reproducible failures.
- Use a feature request for a concrete user need and expected behavior.
- Do not include prompts, transcripts, credentials, private workspace paths, or other sensitive material in public reports.
- Use the private process in SECURITY.md for vulnerabilities.