Documentation

How 8gentHub works

8gentHub is a community for publishing, discovering, exporting, and improving AI agents and skills across ChatGPT, Claude, OpenClaw, Manus, and Perplexity.

14

Core documentation sections

3

Main views: user, creator, and AI integration

1

Searchable reference surface for product behavior

User

How people discover, install, and collaborate

Discover 8gents

Browse the hub, search by keywords, and filter by platform, rating, or preferred model. Each 8gent shows compatibility, creator details, and the latest manifest metadata.

New accounts start with 30 wallet tokens by default, so users can try rewards and installs immediately.

Set preferred models

Preferred models are recommendation tags, not hard locks. They help users understand which model a creator thinks works best for a given platform.

Tokens: charges and earnings

Tokens are used for rewards, creator actions, and AI-assisted installs. Your current wallet balance is available from the Profile page and GET /me/wallet.

How tokens are charged

New accounts start with 30 wallet tokens by default.

Publishing an 8gent charges 5 tokens by default, with discounts available for high-reputation creators.

AI install via POST /ai/recommendations/{recommendation_id}/install charges 5 tokens per install.

Sending a direct creator reward spends exactly the amount you choose.

How tokens are earned

Buy token packs from payments endpoints: Starter 100, Pro 250, Power 700.

Direct creator rewards are split between creator and platform by default: creator 90%, platform 10%.

AI installs are also split between creator and platform. The current default on a 5-token install is creator 4, platform 1.

Publish cost is refunded when an 8gent reaches the run threshold, currently 50 runs.

Helpful discussion contributions grant XP and reputation, not wallet tokens.

Current payout conversion is 700 payout-eligible tokens = $25, with $25 as the minimum request amount.

Collaborate and reward

Tokens create accountability in collaboration. Charging tokens for actions like publishing, direct rewards, and AI installs helps limit spam and low-signal activity so marketplace signals stay meaningful.

Useful reviews, discussions, improvements, and strong 8gent maintenance still matter, but helpful discussion actions now award XP instead of spendable tokens to reduce abuse.

Discussion threads can be archived by creators. Archived threads are hidden by default and become read-only.

Quick Summary

Fast orientation

For users

Browse 8gents, review install outputs, and choose prompt install, npm helper install, or API-led setup.

For creators

Create a reusable 8gent, publish it with metadata and linked skills, then evolve it with versions, discussion, and improvement loops.

For AI integrations

Keep auth modes separate. /ai/* is API-key based, while app and creator routes use Bearer auth.

Creator

How creators publish, improve, and maintain 8gents

Create and publish

Use the Create 8gent flow to generate a draft, edit the manifest, validate required fields, and publish to the community.

Creator guide

Use this workflow to keep your 8gent high-quality and easy to collaborate on.

Recommended creator flow

1. Publish a clean baseline manifest and keep version notes in your thread titles.

2. Review Discussion ideas weekly and convert accepted ideas into improvement proposals.

3. Use archived status to hide old threads from default view without deleting history.

4. Closed idea threads lock improvement form edits by design.

5. For accepted changes, use GitHub contribution with draft PR to compare manifest diffs before merge.

AI Guard and public listing approval flow

1. When you submit an 8gent for public visibility, it enters pending_review.

2. AI Guard evaluates publish-intent metadata and returns a decision for moderation assist.

3. If AI Guard allows the 8gent and account safeguards pass, it may auto-promote to published.

4. If safeguards or risk thresholds trigger, it stays in manual moderation until admin review.

5. Creator-visible states are private, pending_review, published, and rejected.

6. AI Guard scanned 8gents display an AI Guard badge icon in cards.

Moderation and safety notes

Archived threads are read-only, with comments, votes, and proposal edits disabled.

Closed idea threads lock the improvement proposal form.

Removing a published 8gent requires typed confirmation: delete.

Global skills

8gents can link reusable global skills. Each skill has its own slug, category, visibility, version history, and versioned files such as SKILL.md, templates, examples, and checklists.

Creator workflow

1. Open an 8gent and use the Linked Global Skills panel.

2. Link an existing global skill or create a new one directly from the 8gent.

3. New global skills are saved as private by default.

4. New skill versions are saved as private drafts by default.

5. Submit a version explicitly for review with POST /skills/{skill_id}/versions/{version_id}/submit-review.

6. A skill becomes publicly visible only when the skill is public and the relevant version is approved.

7. Install and export packages automatically embed linked skills into the exported SKILL.md entrypoint.

Compatibility notes

Standard installs embed linked global skills into the exported SKILL.md entrypoint.

Single-skill installs use the same single-file SKILL.md shape for compatibility.

Legacy manifest.json.skills is deprecated. Linked global skills are the active model.

Security screening and moderation

Skill metadata and version files are screened for high-risk patterns during submission.

Unsafe submissions are blocked before version creation.

Draft versions remain private until creators submit them for moderation.

Non-admin users can only link approved public skill versions.

Skill moderation endpoints include GET /admin/skills/moderation, POST /admin/skills/versions/{version_id}/approve, and POST /admin/skills/versions/{version_id}/reject.

AutoResearch

AutoResearch helps creators run bounded, self-hosted improvement loops and import results back into 8gentHub.

Install-gated guided flow

AutoResearch appears only when Include AutoResearch setup package in export ZIP is checked in the Install section.

Wizard steps: choose goal, pick benchmark refs, choose starter config, review generated setup, download self-hosted bundle, run locally, import results, review campaign.

There is currently no managed one-click AutoResearch execution endpoint. Self-hosted run plus import is the supported method.

Presets, benchmarks, and prerequisites

Starter presets are Quick smoke test, Balanced iteration, and Budget-capped experiment.

Presets populate max_wall_time_minutes, max_experiments, early_stop_no_gain_runs, max_budget_usd, and notes, while remaining editable.

If no benchmark refs exist on the 8gent, the wizard shows a blocker state with a CTA to add benchmark refs first.

Benchmarks also power public logic pages, manual evaluation runs, leaderboards, and install/export benchmark files.

Self-hosted import and campaign review

POST /agents/{agent_id}/improvements/setup/archive

POST /agents/{agent_id}/improvements/campaigns/import

GET /agents/{agent_id}/improvements/campaigns

DELETE /improvements/campaigns/{campaign_id}

Archive contents include config/autoresearch.config.json, runner/local_autoresearch.py, runner/run-autoresearch.ps1, and runner/run-autoresearch.sh.

Typical sequence: edit config, run locally, review output/autoresearch-results.json, then import it back into 8gentHub.

Manage versions

8gent pages support version history so creators can evolve prompts, tools, and platform-specific settings without losing the previous state of a release.

Benchmark logic note

The benchmark logic page only shows benchmark suites currently saved on the 8gent manifest.

Selecting a benchmark in the editor without saving does not make it available on the logic page yet.

If a requested suite is not saved for that 8gent, the logic page shows the currently available benchmark links instead.

A saved benchmark can still be useful without AutoResearch because it can be used for manual evaluation runs and public leaderboard submissions.

GitHub integration

Supported 8gents can connect to GitHub so maintainers can track repository activity and keep the published workflow aligned with its source project.

AI

References for integrations, exports, and install APIs

AI documents

Start here for AI-to-API integration references, then use the dedicated sections below for API skill export and AI install details.

AI Integration Guide

Open AI Integration Guide

Includes auth matrix, route guardrails, standalone API skill export flow, and platform import expectations.

AI User/Creator Master Prompt

Open Master Prompt

Copy-ready prompt for running recommendation, install, and improvement flows end-to-end.

API: 8gent API Skill

Export a standalone reusable API skill package that is separate from marketplace 8gent installs.

Package shape

Required: SKILL.md.

Recommended: TOOLS.md.

Runtime metadata only where required. ChatGPT exports include openai.yaml.

AGENTS.md is intentionally not included.

Export endpoints: POST /ai/skill/export and POST /ai/skill/export/archive.

Both routes require X-API-Key.

API: AI install

Install APIs export platform-ready artifacts for ChatGPT, Claude, OpenClaw, Manus, and Perplexity. Linked global skills are included automatically in install output, and AI recommendation results expose linked-skill metadata before install.

Install options are intentionally split: download/export ZIP for package contents, copy a concrete install prompt for AI-assisted placement, or use the local npm helper command to copy files from a ZIP or extracted folder.

Install endpoints

POST /ai/recommendations

POST /ai/recommendations/{recommendation_id}/install

POST /agents/{agent_id}/install/validate

POST /agents/{agent_id}/install

POST /agents/{agent_id}/install/archive

Auth matrix

/ai/* uses X-API-Key.

/agents/{agent_id}/install* uses Authorization: Bearer <access_token>.

Wrong: POST /agents/2/install with X-API-Key returns 401.

Right: use the same endpoint with Authorization: Bearer <access_token>.

Install notes

Current AI-driven install cost is 5 tokens per call to POST /ai/recommendations/{recommendation_id}/install.

Recommendation items include linked_skills so AI clients can inspect reusable skill metadata before choosing an install target.

Install responses include linked global skills merged into the exported SKILL.md entrypoint.

Use api.8genthub.com for API calls and keep route families separate by auth mode.

API: improvements

Improvement APIs support self-hosted AutoResearch workflows and discussion-to-GitHub workflows.

Improvement endpoints

POST /agents/{agent_id}/improvements/setup/archive

POST /agents/{agent_id}/improvements/campaigns/import

GET /agents/{agent_id}/improvements/campaigns

DELETE /improvements/campaigns/{campaign_id}

POST /agents/{agent_id}/discussions/{thread_id}/improvement-proposal

POST /agents/{agent_id}/discussions/{thread_id}/improvement-accept

POST /agents/{agent_id}/discussions/{thread_id}/improvement-reject

POST /agents/{agent_id}/discussions/{thread_id}/github/contribute

Use mode=issue_plus_pr_draft to open a draft PR that includes synced manifest.json diffs.

Need quick answers?

Visit the FAQ for common publishing, installation, and account questions.