Resolvent

A world-model repository and a generic research plugin by Autonomio.

Resolvent helps you research and refine questions in chat, read your world model for context or answers, and explicitly add selected findings, questions or discussion. One plugin serves compatible GitHub repositories through your authorized connection. Invoking it alone authorizes neither model reads nor writes. Contributions preserve evidence and uncertainty through admission, independent review, CI and authorized merge.

A claim's semantic core is a short self-contained statement, categorical uncertainty and an evidence record. IDs, class and lifecycle are bookkeeping. This template contains no world-model claims or inherited research data.

Start a world model

1. Create a repository from this template. Choose its owner and visibility.

2. Install Resolvent from the ChatGPT or Claude directory once published. Use the authorized GitHub connection available in the invoking chat.

3. Select your destination together with your question:

   @resolvent OWNER/REPOSITORY Research [my question] and update the world model.

For work already in the chat:

@resolvent OWNER/REPOSITORY Capture the useful findings from this conversation. Do not do new research.

Name a repository or organization when model access is needed. Resolvent verifies the destination, asks if ambiguous, and retains it in the current chat. An optional native project instruction can reuse the selection across chats. Chat-only research and reformulation need no destination. Examples and setup. A full contribution finishes with verified merge under the user's actual authorization and repository rules; explicit preflight, submit-only, review-only and no-merge limits prevail.

Plugin releases and distributions

Plugin 1.4.1 aligns contribution guidance with core 1.3.1's 10 MiB per-data-file limit. Existing worlds adopt the larger limit through their normal core update. Plugin 1.4.0 adds a public MCP verifier for current workflow delivery and protocol compatibility to the fixed client and separately versioned live workflow references, including explicit read/write intent, distillation and integrated question reformulation. World-model protocol 1.2 adds optional reviewed source ancestry, retaining legacy 1.0/1.1 inputs. Read-only lineage reports expose possible common origins without promoting independence or uncertainty. Its developer is Autonomio and its plugin website is https://resolvent.autonom.io. After a reviewed main-branch change passes validation, CI publishes versioned GitHub assets and the standalone plugin-release branch. Public directory setup is a separate first step.

• Latest published packages
• ChatGPT plugin ZIP
• Claude plugin ZIP
• Plugin release and compatibility instructions

Both ZIPs contain identical code, skills, assets and synchronized host manifests. Local catalogs are in .agents/plugins/marketplace.json and .claude-plugin/marketplace.json; their source is the generated standalone plugin at dist/plugin/resolvent/. Claude's public directory can scan the release branch and automatically publish passing versions once enabled. ChatGPT workspace admins can import the GitHub marketplace once and automatically sync plugin-release; setup instructions explain the scope. Public ChatGPT directory skill changes require ZIP upload, review and publication. Data-only updates create no plugin release. The client reads verified operational references from the separate workflow-release branch. A routine workflow update changes that channel without replacing the installed ZIP. Source content is pinned for each operation; fixed permission and destination rules remain installed. This is an application design requiring host verification, not a claim that the public ChatGPT directory polls arbitrary websites for packages. For direct Claude marketplace distribution, the installation line is:

/plugin install resolvent --marketplace autonomio/resolvent

The standalone dist/resolvent.zip skill remains an optional compatibility artifact. A skill/plugin cannot grant GitHub permissions. File-sync-only access cannot create contribution PRs. If needed operations are unavailable, Resolvent provides a truthful incomplete receipt; installation alone is not evidence of research, admission or merge.

Repository setup

Existing repository agents read AGENTS.md and their job instructions. Give independent review agents access. The originating chat owns its selected contribution's authoring and corrections; external review remains independent. No agent provider, model key or organization-hosted service is required by the plugin. The public workflow verifier has no repository access; model operations use the host's existing GitHub authority.

Template copies do not inherit branch protections or app grants. Enable Actions, require World model integrity and Tests and skill package, independent review, stale-approval dismissal, resolved conversations and up-to-date branches. See operations for setup and weekly/manual maintenance.

What lives where

Path	Purpose
skills/resolvent/	Shared generic researcher-facing workflow and references.
workflows/resolvent/	Separately versioned live operational guidance.
plugin.json, .codex-plugin/, .claude-plugin/	Portable and host-compatible plugin metadata.
dist/chatgpt/, dist/claude/, dist/plugin/	Reproducible plugin distributions and local catalog source.
input-artefacts/RUN-*.json	Immutable retained research/capture contributions.
claims/, evidence/, sources/	Canonical assertions, evidence and source anchors.
graph/R-*.yaml	Justified claim-to-claim connectors, stored once.
briefs/, questions/, maintenance/	Research execution, closure, open questions and maintenance.
nightly-briefs/YYYY-MM-DD.md	Reports on completed recent maintenance passes; not knowledge.
docs/contracts/, docs/schemas/, scripts/	Governed processes, executable schemas and deterministic tools.

The skill routes research, conversation, interview, source material, simulation, intra-model reasoning and liminal research. Existing-chat capture does no new topic research. Admission creates justified immediate connections; maintenance discovers broader old-to-new and old-to-old relationships. A no-connection finding is valid.

An optional Portal explores the repository through search and visualization. It is separate from this generic plugin and is not required for contributions.

Maintainers

Follow the plugin compatibility rule in AGENTS.md for every system/package change. Assess whether installed plugin behavior needs updating; ordinary model-data updates alone do not require a plugin release. Build and check both distributions as documented in plugin releases.

Resolvent software and the plugin use the MIT License. Organization world-model data and third-party source material are excluded and retain their rights holders' terms. Forking or updating the software grants no rights to that data.