🛡️ Endor Labs Agent Kit

Ready-to-use Endor Labs agents for AI coding assistants, plus the recipe-first builder used to maintain and publish them.

Tip

Pick a lane first. The repository is both an installable agent catalog and the source generator for the public endorlabs/ai-plugins mirror.

Start Here

I want to...	Go here
🚀 Install agents in a coding assistant	docs/getting-started.md or the host package README
🧰 Contribute or propose an agent	docs/contributing-agents.md
🤖 Ask an agent to inspect or sync this repo	docs/for-agents.md
🧰 Change how agents are generated	docs/maintainer-guide.md
📦 Publish or mirror to ai-plugins	docs/distribution-sync.md
🧱 Embed the workflows in another runtime	docs/portable-runtime-conformance.md

A machine-readable index is available in llms.txt.

🧭 Browse The Kit

Area	What is inside	Start here
🤖 Plugin agents	Host packages for Claude Code, Codex, Gemini CLI, Antigravity CLI, and Cursor	plugins/README.md
🖱️ Cursor IDE	Cursor plugin metadata, generated agents, support skills, and advisory hooks	.cursor-plugin/plugin.json
🐍 Cursor SDK	Python SDK launcher and generated prompts for automation	cursor-sdk/README.md
🧩 Single-agent bundles	Manual per-host artifacts and README files	Agent Catalog
🧱 Portable bundles	Runtime-neutral prompts, manifests, output contracts, and adapter expectations	portable/
🛠️ Source recipes	Maintainer-only source of truth for behavior, evals, actions, and diagrams	source/agents/
🔒 Guardrails	Safety, provenance, runtime, and documentation validation	docs/guardrails.md

Table Of Contents

• 🚀 Plugin Quick Start
• 🐍 Cursor SDK Quick Start
• Agent Quick Start
• 🧩 Capabilities And Skills
• 🔒 Boundaries And Rules
• 🤖 Agent Operating Rules
• 📇 Agent Catalog
• 🗂️ Which Directory Do I Use?
• 🖥️ Supported Hosts
• 🧱 Already Have Your Own Tech Stack Or Workflows Wired?
• 🔌 MCP Usage
• 📦 Plugin Packaging Route
• ✅ Release Checklist
• 🧪 Output Contract
• 🛡️ Safety Model
• 🧰 Contribute An Agent
• 📖 Reference

🚀 Plugin Quick Start

Current generated plugin package version: 2.1.0.

Host	Best for	First move
Claude Code	Claude Code plugin agents, setup skill, and advisory hooks	Read plugins/claude/endor-labs-agent-kit/README.md, then install endor-labs-agent-kit@endorlabs.
Codex	Skills plus optional bundled custom-agent TOML files	Read plugins/codex/endor-labs-agent-kit/README.md.
Gemini CLI	Extension with skills and preview subagents	Read plugins/gemini/endor-labs-agent-kit/README.md.
Antigravity CLI	Plugin with skills and subagents	Read plugins/antigravity/endor-labs-agent-kit/README.md.
Cursor IDE	Customer-facing Cursor plugin agents	Install from .cursor-plugin/, root agents/, root skills/, root hooks/, and assets/logo.png.

Google documents Antigravity CLI as the consumer transition path for Gemini CLI. Use the Gemini package for supported Gemini CLI environments and compatibility checks; use the Antigravity package for affected Gemini CLI consumer accounts.

After installing any host package, run setup first:

Use the endor-agent-kit-setup skill to check Endor Agent Kit readiness. Do not run scans.

The plugin packages are the lowest-friction way to load Endor Labs workflows into Claude Code, Codex, Gemini CLI, Antigravity CLI, or Cursor. They package setup guidance and generated workflow agents/skills from the same source recipes as the manual catalog without injecting every recipe into the active model context.

Claude compatibility note: ai-plugins@endorlabs remains available for existing Claude Code users and pinned installs. New users should prefer endor-labs-agent-kit@endorlabs. Do not enable both Claude plugin ids in one profile because they expose the same setup skill and agents. The plugin does not auto-disable, uninstall, or edit Claude settings for either id.

After installing a plugin, ask the host to use the endor-agent-kit-setup skill first. Setup checks local readiness, guides endorctl authentication and namespace selection, reports gh and toolchain gaps, and offers host-specific self-checks before live Endor lookups.

🐍 Cursor SDK Quick Start

Use cursor-sdk/ when you want to launch Cursor agents from Python code, CI, orchestration, or a backend service. Use the Cursor plugin package when you want agents installed into the Cursor IDE.

python3 -m pip install -r cursor-sdk/requirements.txt
export CURSOR_API_KEY="crsr_..."
python cursor-sdk/run_cursor_agent.py endor-probe-droid-agent \
  --workspace /path/to/repo \
  "Explain what evidence you need to assess GitHub onboarding gaps. Keep it read-only."

The SDK package includes cursor-sdk/agent_definitions.json, generated prompt files under cursor-sdk/agents/, and a runnable run_cursor_agent.py launcher for local or cloud agents.

Agent Quick Start

Agent	What it does	Invoke	Safety	First prompt
🔎 AI SAST Triage	Triage Endor AI SAST findings, use exploit and remediation context, and open requested change requests	endor-ai-sast-triage-agent	approval-gated mutating	Triage AI SAST findings for this repository. Do not edit files, open a PR/MR, create a ticket, or write an Endor policy until I approve the specific gate.
🧭 CI/CD And Supply Chain Posture	Assess CI/CD and supply chain posture from existing Endor findings and read-only GitHub configuration evidence	endor-cicd-posture-agent	read-only	Assess CI/CD and supply chain posture for namespace <namespace>. Keep it read-only and validate the deterministic score.
⚖️ Dependency Decision Helper	Decide whether to add, upgrade to, or keep a specific package version	endor-dependency-decision-helper-agent	read-only	Assess whether we should use npm lodash version 4.17.20. Keep it read-only.
📊 Endor Labs Package Risk Summary	Summarize the risk profile of a specific package version	endor-package-risk-summary-agent	read-only	Summarize npm lodash version 4.17.20 with verified Endor evidence. Keep it read-only.
📚 Endor Labs Repository Dependency Reviewer	Review local dependency manifests with read-only file inspection and Endor evidence	endor-repository-dependency-reviewer-agent	read-only	Review this repository's dependency manifests with read-only evidence only.
⬆️ Endor Labs Upgrade Impact Analysis	Analyze Endor platform upgrade impact with VersionUpgrade, CIA, findings, and manifest context	endor-upgrade-impact-analysis-agent	read-only	Show the safest upgrade path for repository <owner>/<repo> package lodash. Keep it read-only.
💬 Endor Labs Vulnerability Explainer	Understand a specific CVE, GHSA, or Endor vulnerability and what to do next	endor-vulnerability-explainer-agent	read-only	Explain CVE-2021-44228 using verified Endor evidence. Keep it read-only.
🧯 Endor Troubleshooter	Diagnose Endor Labs errors, warnings, scan failures, slow scans, missing integrations, SSO, containers, policy, and reachability issues	endor-troubleshooter-agent	read-only	Diagnose this Endor issue from redacted error text and read-only tenant evidence. Keep it read-only.
🔍 Findings Browser	Browse, filter, and summarize existing Endor findings with read-only namespace-scoped queries	endor-findings-browser-agent	read-only	Show the critical and high reachable findings for namespace <namespace>. Keep it read-only.
🤖 Malware Response Agent	Use an Endor Labs workflow agent	endor-malware-response-agent	read-only	Use the malware-response workflow. Keep it within its generated safety contract.
📡 Probe Droid	Probe GitHub.com onboarding gaps and prescribe Endor scan profiles, toolchains, package integrations, and reachability setup	endor-probe-droid-agent	read-only	Explain what evidence you need to assess GitHub onboarding gaps for this repository. Keep it read-only.
🗺️ Remediation Planner	Preview safe dependency remediation options without opening PRs	endor-remediation-planner-agent	read-only	Preview remediation options for this repository. Do not edit files or open a PR/MR.
🛠️ SCA Remediation	Remediate dependency vulnerabilities with Endor SCA findings, UIA evidence, low-risk PR lanes, deterministic risk decisions, validation, and approved PR/MR creation	endor-sca-remediation-agent	approval-gated mutating	Inspect this repository and prepare a remediation plan only. Do not edit files, create branches, push, open a PR/MR, create a ticket, or write Endor policy.

🧩 Capabilities And Skills

When a plugin package is loaded, start from the user job and let the host map to the generated skill or agent name.

Job	Claude Code	Codex	Gemini CLI	Antigravity CLI	Cursor
Set up this machine and self-check readiness	Skill endor-agent-kit-setup	Skill endor-agent-kit-setup	Skill endor-agent-kit-setup	Skill endor-agent-kit-setup	Agent endor-agent-kit-setup-agent, skill endor-agent-kit-setup
Triage Endor AI SAST findings	Agent ai-sast-triage	Skill ai-sast-triage, custom agent endor-ai-sast-triage-agent	Skill/subagent ai-sast-triage	Skill/subagent ai-sast-triage	Agent endor-ai-sast-triage-agent, skill ai-sast-triage
Assess CI/CD and supply chain posture	Agent cicd-posture	Skill cicd-posture, custom agent endor-cicd-posture-agent	Skill/subagent cicd-posture	Skill/subagent cicd-posture	Agent endor-cicd-posture-agent, skill cicd-posture
Diagnose Endor setup, scan, or integration issues	Agent endor-troubleshooter	Skill endor-troubleshooter, custom agent endor-troubleshooter-agent	Skill/subagent endor-troubleshooter	Skill/subagent endor-troubleshooter	Agent endor-troubleshooter-agent, skill endor-troubleshooter
Assess GitHub onboarding gaps	Agent probe-droid	Skill probe-droid, custom agent endor-probe-droid-agent	Skill/subagent probe-droid	Skill/subagent probe-droid	Agent endor-probe-droid-agent, skill probe-droid
Find safe SCA remediation paths	Agent sca-remediation	Skill sca-remediation, custom agent endor-sca-remediation-agent	Skill/subagent sca-remediation	Skill/subagent sca-remediation	Agent endor-sca-remediation-agent, skill sca-remediation
Evaluate package risk or a dependency decision	Agents dependency-decision-helper, package-risk-summary	Skills dependency-decision-helper, package-risk-summary; custom agents endor-dependency-decision-helper-agent, endor-package-risk-summary-agent	Skills/subagents dependency-decision-helper, package-risk-summary	Skills/subagents dependency-decision-helper, package-risk-summary	Agents endor-dependency-decision-helper-agent, endor-package-risk-summary-agent; skills dependency-decision-helper, package-risk-summary
Review repository dependencies or remediation options	Agents repository-dependency-reviewer, remediation-planner	Skills repository-dependency-reviewer, remediation-planner; custom agents endor-repository-dependency-reviewer-agent, endor-remediation-planner-agent	Skills/subagents repository-dependency-reviewer, remediation-planner	Skills/subagents repository-dependency-reviewer, remediation-planner	Agents endor-repository-dependency-reviewer-agent, endor-remediation-planner-agent; skills repository-dependency-reviewer, remediation-planner
Analyze upgrade impact or explain vulnerabilities	Agents upgrade-impact-analysis, vulnerability-explainer	Skills upgrade-impact-analysis, vulnerability-explainer; custom agents endor-upgrade-impact-analysis-agent, endor-vulnerability-explainer-agent	Skills/subagents upgrade-impact-analysis, vulnerability-explainer	Skills/subagents upgrade-impact-analysis, vulnerability-explainer	Agents endor-upgrade-impact-analysis-agent, endor-vulnerability-explainer-agent; skills upgrade-impact-analysis, vulnerability-explainer

The full catalog below still matters for manual installs, portable bundles, and future agent contributions.

🔒 Boundaries And Rules

• Always preserve the generated recipe safety contract, approval gates, and evidence requirements.
• Always treat setup as readiness and configuration guidance; setup must not run scans or mutate repositories.
• Always report namespace, authentication, MCP, endorctl, gh, or toolchain gaps before live Endor work.
• Never print, persist, or copy secret values; report credential presence only by variable or key name.
• Never run endorctl host-check from setup or generated plugin guidance.
• Never claim a file edit, branch push, PR/MR, ticket, policy write, or approval unless the host or adapter returned evidence.

🤖 Agent Operating Rules

If you are an AI agent reading this repository:

1. Decide whether the user is installing, reviewing, editing, or publishing.
2. For installs, read the selected generated README and copy the generated artifact exactly. Keep sibling files such as architecture.svg, actions.yaml, and endorctl-setup.md with the artifact when present.
3. For behavior changes, edit source/agents/<agent>/recipe.yaml, instructions.md, optional actions.yaml, evals, or publication templates. Do not hand-edit generated customer-facing copies as the source of truth.
4. Regenerate with endor-agent-kit publish source/agents/*/recipe.yaml --dest . --prune --include-plugins when plugin packages are in scope.
5. Validate with tests, guardrails, provenance checks, generated-artifact drift checks, and the provider checks listed in docs/plugin-release-checklist.md.
6. Sync ai-plugins only after Agent Kit generation is clean; use docs/distribution-sync.md for the mirror boundary.

📇 Agent Catalog

Generated artifacts are checked in so users can copy or install agents without running the builder. Maintainer source recipes live under source/agents/ and are the maintainer-facing source of truth.

If you are installing an agent, start with the generated host directories below. You only need source/agents/ when you are changing or contributing an agent.

Agent	Use it when you want to...	Claude Code	Claude Managed Agents	Codex	Gemini	Cursor	Cursor SDK	Portable
AI SAST Triage	Triage Endor AI SAST findings, use exploit and remediation context, and open requested change requests	claude-code/ai-sast-triage/	-	codex/ai-sast-triage/	gemini/ai-sast-triage/	agents/endor-ai-sast-triage-agent.md + skills/ai-sast-triage/	cursor-sdk/agents/endor-ai-sast-triage-agent.md	portable/ai-sast-triage/
CI/CD And Supply Chain Posture	Assess CI/CD and supply chain posture from existing Endor findings and read-only GitHub configuration evidence	claude-code/cicd-posture/	claude-managed-agents/cicd-posture/	codex/cicd-posture/	gemini/cicd-posture/	agents/endor-cicd-posture-agent.md + skills/cicd-posture/	cursor-sdk/agents/endor-cicd-posture-agent.md	portable/cicd-posture/
Dependency Decision Helper	Decide whether to add, upgrade to, or keep a specific package version	claude-code/dependency-decision-helper/	claude-managed-agents/dependency-decision-helper/	codex/dependency-decision-helper/	gemini/dependency-decision-helper/	agents/endor-dependency-decision-helper-agent.md + skills/dependency-decision-helper/	cursor-sdk/agents/endor-dependency-decision-helper-agent.md	portable/dependency-decision-helper/
Endor Labs Package Risk Summary	Summarize the risk profile of a specific package version	claude-code/package-risk-summary/	claude-managed-agents/package-risk-summary/	codex/package-risk-summary/	gemini/package-risk-summary/	agents/endor-package-risk-summary-agent.md + skills/package-risk-summary/	cursor-sdk/agents/endor-package-risk-summary-agent.md	portable/package-risk-summary/
Endor Labs Repository Dependency Reviewer	Review local dependency manifests with read-only file inspection and Endor evidence	claude-code/repository-dependency-reviewer/	-	codex/repository-dependency-reviewer/	gemini/repository-dependency-reviewer/	agents/endor-repository-dependency-reviewer-agent.md + skills/repository-dependency-reviewer/	cursor-sdk/agents/endor-repository-dependency-reviewer-agent.md	portable/repository-dependency-reviewer/
Endor Labs Upgrade Impact Analysis	Analyze Endor platform upgrade impact with VersionUpgrade, CIA, findings, and manifest context	claude-code/upgrade-impact-analysis/	claude-managed-agents/upgrade-impact-analysis/	codex/upgrade-impact-analysis/	gemini/upgrade-impact-analysis/	agents/endor-upgrade-impact-analysis-agent.md + skills/upgrade-impact-analysis/	cursor-sdk/agents/endor-upgrade-impact-analysis-agent.md	portable/upgrade-impact-analysis/
Endor Labs Vulnerability Explainer	Understand a specific CVE, GHSA, or Endor vulnerability and what to do next	claude-code/vulnerability-explainer/	claude-managed-agents/vulnerability-explainer/	codex/vulnerability-explainer/	gemini/vulnerability-explainer/	agents/endor-vulnerability-explainer-agent.md + skills/vulnerability-explainer/	cursor-sdk/agents/endor-vulnerability-explainer-agent.md	portable/vulnerability-explainer/
Endor Troubleshooter	Diagnose Endor Labs errors, warnings, scan failures, slow scans, missing integrations, SSO, containers, policy, and reachability issues	claude-code/endor-troubleshooter/	claude-managed-agents/endor-troubleshooter/	codex/endor-troubleshooter/	gemini/endor-troubleshooter/	agents/endor-troubleshooter-agent.md + skills/endor-troubleshooter/	cursor-sdk/agents/endor-troubleshooter-agent.md	portable/endor-troubleshooter/
Findings Browser	Browse, filter, and summarize existing Endor findings with read-only namespace-scoped queries	claude-code/findings-browser/	claude-managed-agents/findings-browser/	codex/findings-browser/	gemini/findings-browser/	agents/endor-findings-browser-agent.md + skills/findings-browser/	cursor-sdk/agents/endor-findings-browser-agent.md	portable/findings-browser/
Malware Response Agent	Use an Endor Labs workflow agent	claude-code/malware-response/	claude-managed-agents/malware-response/	codex/malware-response/	gemini/malware-response/	agents/endor-malware-response-agent.md + skills/malware-response/	cursor-sdk/agents/endor-malware-response-agent.md	portable/malware-response/
Probe Droid	Probe GitHub.com onboarding gaps and prescribe Endor scan profiles, toolchains, package integrations, and reachability setup	claude-code/probe-droid/	claude-managed-agents/probe-droid/	codex/probe-droid/	gemini/probe-droid/	agents/endor-probe-droid-agent.md + skills/probe-droid/	cursor-sdk/agents/endor-probe-droid-agent.md	portable/probe-droid/
Remediation Planner	Preview safe dependency remediation options without opening PRs	claude-code/remediation-planner/	-	codex/remediation-planner/	gemini/remediation-planner/	agents/endor-remediation-planner-agent.md + skills/remediation-planner/	cursor-sdk/agents/endor-remediation-planner-agent.md	portable/remediation-planner/
SCA Remediation	Remediate dependency vulnerabilities with Endor SCA findings, UIA evidence, low-risk PR lanes, deterministic risk decisions, validation, and approved PR/MR creation	claude-code/sca-remediation/	-	codex/sca-remediation/	gemini/sca-remediation/	agents/endor-sca-remediation-agent.md + skills/sca-remediation/	cursor-sdk/agents/endor-sca-remediation-agent.md	portable/sca-remediation/

🗂️ Which Directory Do I Use?

Goal	Start Here	You Do Not Need
Install the Claude Code plugin package	plugins/claude/endor-labs-agent-kit/README.md or legacy plugins/claude/ai-plugins/README.md	source/, src/, tests/
Install the Codex plugin package	plugins/codex/endor-labs-agent-kit/README.md	source/, src/, tests/
Install the Gemini CLI extension package	plugins/gemini/endor-labs-agent-kit/README.md	source/, src/, tests/
Install the Antigravity CLI plugin package	plugins/antigravity/endor-labs-agent-kit/README.md	source/, src/, tests/
Install the Cursor package	.cursor-plugin/, root agents/, root skills/, root hooks/, and assets/logo.png	Gemini extension files, source/, src/, tests/
Run Cursor SDK automation	cursor-sdk/README.md	Cursor IDE plugin metadata, Gemini extension files, source/
Install a Claude Code agent	claude-code/<agent>/README.md	source/, src/, tests/
Install a Claude Managed Agent	claude-managed-agents/<agent>/README.md	source/, src/, tests/
Install a Codex skill	codex/<agent>/README.md	source/, src/, tests/
Install a Gemini CLI skill and subagent	gemini/<agent>/README.md	source/, src/, tests/
Install a portable runtime-neutral bundle	portable/<agent>/README.md	source/, src/, tests/
Modify or contribute an agent	source/agents/<agent>/recipe.yaml and instructions.md	Generated catalog files as the first edit
Work on the kit builder itself	src/endor_agent_kit/ and tests/	Host install directories unless compiler output changes

The source/agents/ tree is for maintainers and contributors. It is not copied into Claude Code or Managed Agents. Installable artifacts are the generated host directories listed in the catalog.

🖥️ Supported Hosts

Host	Generated path	Typical install target
Claude Code	plugins/claude/endor-labs-agent-kit/, legacy plugins/claude/ai-plugins/, and claude-code/<agent>/	Claude Code plugin marketplace, or .claude/agents/ for manual per-agent installs
Claude Managed Agents	claude-managed-agents/<agent>/	Anthropic Console or ant CLI agent and environment creation
Codex	plugins/codex/endor-labs-agent-kit/ and codex/<agent>/	Codex plugin marketplace, bundled global custom agents, or $HOME/.agents/skills/<agent>/ for manual skill installs
Gemini	plugins/gemini/endor-labs-agent-kit/ and gemini/<agent>/	Gemini CLI extension install, or manual skill/subagent reference from gemini/<agent>/
Antigravity	plugins/antigravity/endor-labs-agent-kit/	Antigravity CLI plugin install with generated skills and subagents
Cursor	.cursor-plugin/, agents/<agent>.md, skills/<agent>/, hooks/, and assets/logo.png	Cursor plugin install with generated agents, support skills, and advisory hooks; Gemini extension files are separate
Cursor SDK	cursor-sdk/	Python automation, CI, orchestration, backend services, local SDK agents, or Cursor cloud agents
Portable	portable/<agent>/	Customer-managed agent runtime, workflow engine, or internal platform

🧱 Already Have Your Own Tech Stack Or Workflows Wired?

Use the portable/<agent>/ bundles when your organization already has an agent runtime, repository workflow, ticketing workflow, approval system, credential controls, and audit pipeline. Portable bundles give you the agent instructions and runtime contract without assuming Claude Code, Claude Managed Agents, Codex, Gemini, Antigravity, Cursor, Cursor SDK, or any other host-specific package shape.

Each portable bundle includes:

• agent.md: the generated runtime-neutral agent instructions.
• agent.manifest.json: machine-readable transports, capabilities, actions, wrappers, and degradation behavior.
• output-contract.md: inputs, outputs, adapter contracts, and mechanical workflow gates when available.
• Optional actions.yaml portable adapter contracts plus endorctl-setup.md and architecture.svg support files.

The runtime integration model is deliberately split:

• Agent Kit defines the workflow, evidence requirements, safety contract, and structured output.
• Your runtime enforces authentication, authorization, logging, audit, adapter execution, and approval policy.
• Runtime adapters perform semantic actions such as endor.query, source.change_request.create, approval.verify, endor.policy.write, and ticket.create.
• The agent must not claim an action completed unless the runtime adapter returns evidence such as a PR/MR URL, ticket ID, policy UUID, branch, validation result, or explicit data gap.

Example adapter mappings:

Portable action	Example runtime adapters
endor.query	Endor API proxy, endorctl api, approved Endor MCP adapter
source.change_request.create	GitHub pull request, GitLab merge request, Bitbucket pull request, internal change workflow
ticket.create	Jira issue, ServiceNow task, Linear issue, internal ticketing
approval.verify	AppSec approval service, source-provider approval API, internal risk-acceptance workflow

Example runtime wrapper:

System:
Load portable/<agent>/agent.md as the generated instruction source.
Expose only the adapters allowed by portable/<agent>/agent.manifest.json and your organization policy.
Pause for confirmation when an action declares confirmation_required=true.
Return adapter evidence or a structured data gap after every action.

User:
Use this agent to analyze repository <repo>. Prefer ticket creation over a source change request unless the plan is low risk and validation evidence is available.

For remediation workflows, let the agent find the right remediation path first. At the mutation gate, your runtime can offer approved targets such as plan-only output, source change request creation, ticket creation, or both. Remediation workflows may declare ticket.create directly; other portable bundles can still use ticket.create as a runtime wrapper after final output. The agent only claims ticket creation when the runtime performs it and returns ticket evidence.

Portable bundles are generated artifacts. Configure local adapters and organization policy outside the bundle. If agent behavior needs to change, edit the Source Recipe and regenerate the catalog.

🔌 MCP Usage

MCP is not used by the mutating remediation workflows. AI SAST Triage, SCA Remediation, Remediation Planner, Upgrade Impact Analysis, Probe Droid, and the Codex skills use documented Endor API or endorctl api paths instead.

MCP remains in the catalog only where the current public recipe still depends on Endor package/vulnerability lookup tools that do not yet have an endorctl api contract in this kit:

Agent	MCP use	Non-MCP path in same artifact
Dependency Decision Helper	Package risk, vulnerability list, and vulnerability enrichment.	endorctl api for package scores, license, and similar-package signals.
Endor Labs Package Risk Summary	Package risk, vulnerability list, and vulnerability enrichment.	endorctl api for package scores, license, and similar-package signals.
Endor Labs Repository Dependency Reviewer	Per-dependency risk and vulnerability checks after local read-only manifest inspection.	No non-MCP path currently.
Endor Labs Vulnerability Explainer	Vulnerability detail lookup.	No non-MCP path currently.

If MCP is unavailable, those agents must record the missing signal in data_gaps rather than blocking install or fabricating evidence.

Portable bundles do not configure MCP servers. They declare transport requirements in agent.manifest.json. If a portable agent requires MCP, the customer runtime must provide an MCP-compatible Endor adapter exposing the required tools; otherwise the agent records the missing signal in data_gaps.

📦 Plugin Packaging Route

Claude Code and Codex still publish generated artifacts under claude-code/<agent>/ and codex/<agent>/ for manual installs. Plugin package generation is available as an opt-in publication slice:

endor-agent-kit publish source/agents/*/recipe.yaml --dest . --prune --include-plugins

Generated plugin packages currently include:

• plugins/codex/endor-labs-agent-kit/: Codex plugin skills, bundled custom-agent TOML files, setup skill, installer script, and Codex marketplace metadata under .agents/plugins/marketplace.json and plugins/codex/.agents/plugins/marketplace.json.
• plugins/claude/endor-labs-agent-kit/: Claude Code plugin agents, setup skill, and Claude marketplace metadata under .claude-plugin/marketplace.json plus plugins/claude/.claude-plugin/marketplace.json for package-local testing.
• plugins/claude/ai-plugins/: legacy Claude Code compatibility package for existing ai-plugins@endorlabs installs. New installs should prefer endor-labs-agent-kit@endorlabs; do not enable both ids in one Claude profile.
• plugins/gemini/endor-labs-agent-kit/: Gemini CLI extension with setup skill, Gemini workflow skills, preview subagents, minimal context, and no zip release artifact.
• plugins/antigravity/endor-labs-agent-kit/: Antigravity CLI plugin with setup skill, Antigravity workflow skills, subagents, minimal assets, and a root plugin.json.
• .cursor-plugin/ plus root agents/, skills/, hooks/, and assets/logo.png: Cursor plugin metadata, generated Cursor workflow agents, setup agent, support skills, and advisory hooks. Cursor does not install Gemini CLI extension files; use plugins/gemini/endor-labs-agent-kit/ for Gemini.
• Root .mcp.json and GEMINI.md: optional MCP support context for the multi-host repo root. The root must not contain gemini-extension.json.
• cursor-sdk/: Cursor Python SDK automation package with generated prompt files, agent_definitions.json, requirements.txt, and a runnable run_cursor_agent.py launcher for local or cloud SDK agents.

All plugin packages preserve the same recipe source, action metadata, and approval gates as the manual generated catalog. Cursor package files stay at the repository root because the public Cursor package source is ./. Gemini installs from the generated extension directory for local validation or from the tagged GitHub repository for public distribution. Antigravity installs from the generated plugin directory. No zip artifact is generated for these packages. See docs/plugin-packaging-design.md for blast-radius notes.

✅ Release Checklist

Use docs/plugin-release-checklist.md before tagging or publishing plugin packages. It records the provider-specific publish paths, local validation commands, public GitHub distribution steps, and external documentation freshness checks for Claude Code, Codex, Gemini, Antigravity, Cursor, Cursor SDK, and Endor API/docs context.

Editions

Most users should not need to think about editions. The current catalog publishes one customer-facing artifact per agent and host, directly under claude-code/<agent>/, claude-managed-agents/<agent>/, codex/<agent>/, gemini/<agent>/, cursor-sdk/agents/<agent>.md, or portable/<agent>/.

The builder still understands internal developer-edition and enterprise-edition recipe sections for compatibility and for future agents that genuinely need multiple artifacts. When a recipe selects one artifact, the published directory is flat and the generated README omits the edition label.

Shell access is still controlled by each recipe's host capability contract. Read-only agents that do not need endorctl api deny Bash. Read-only agents that need endorctl api allow Bash only for documented read-only Endor lookup commands. Mutating agents keep file edits, branch pushes, PR/MR creation, comments, approval verification, and Endor policy writes behind separate approval gates.

Install An Agent

For Claude Code, Codex, Gemini CLI, or Antigravity CLI, prefer the plugin package README when you want the full generated workflow set and setup guidance. For Cursor, use the .cursor-plugin/ metadata, root agents/, root skills/, root hooks/, and assets/logo.png. For a single agent, pick an agent from the catalog, then open that host directory's README. If the agent has edition subdirectories, choose the one that matches your environment; otherwise use the agent directory directly. For scripted Cursor automation, use cursor-sdk/README.md instead of the Cursor IDE plugin metadata.

Ask An LLM To Install It

If you are using an assistant that can edit files or run commands in your target repository, copy and paste this prompt:

Install this Endor Labs Agent Kit agent in the current repository.

Agent Kit root: /path/to/endor-labs-agent-kit
Host: claude-code
Agent directory: claude-code/ai-sast-triage

Please:
1. Read the install README at <Agent Kit root>/<Agent directory>/README.md.
2. Install the generated agent artifact from that directory into this repository.
3. Preserve the generated agent prompt exactly; do not rewrite or summarize it.
4. Tell me any Endor MCP if declared, endorctl, repository, or credential setup still required.
5. Show me the command or prompt I should use to invoke the agent.

Replace Agent directory with the directory you selected from the catalog.

Claude Code

Copy the generated subagent into your target repository and restart Claude Code if needed.

mkdir -p .claude/agents
cp /path/to/endor-labs-agent-kit/claude-code/ai-sast-triage/ai-sast-triage.md \
  .claude/agents/ai-sast-triage.md

Then invoke it from Claude Code:

@agent-ai-sast-triage triage AI SAST findings for this repository

Claude Managed Agents

Create the agent and environment with the Anthropic CLI or Console. If the selected agent declares MCP in its generated README.md, update the MCP server URL and vault references in the generated YAML first.

cd /path/to/endor-labs-agent-kit/claude-managed-agents/dependency-decision-helper
ant beta:agents create < agent.yaml
ant beta:environments create < environment.yaml

Use session-template.yaml as the starting point when creating sessions.

Codex

Copy the generated skill directory into your Codex skills directory, then start a new Codex session so the skill loader can see it.

mkdir -p "$HOME/.agents/skills"
cp -R /path/to/endor-labs-agent-kit/codex/ai-sast-triage \
  "$HOME/.agents/skills/ai-sast-triage"
cp -R /path/to/endor-labs-agent-kit/codex/cicd-posture \
  "$HOME/.agents/skills/cicd-posture"
cp -R /path/to/endor-labs-agent-kit/codex/dependency-decision-helper \
  "$HOME/.agents/skills/dependency-decision-helper"
cp -R /path/to/endor-labs-agent-kit/codex/package-risk-summary \
  "$HOME/.agents/skills/package-risk-summary"
cp -R /path/to/endor-labs-agent-kit/codex/repository-dependency-reviewer \
  "$HOME/.agents/skills/repository-dependency-reviewer"
cp -R /path/to/endor-labs-agent-kit/codex/upgrade-impact-analysis \
  "$HOME/.agents/skills/upgrade-impact-analysis"
cp -R /path/to/endor-labs-agent-kit/codex/vulnerability-explainer \
  "$HOME/.agents/skills/vulnerability-explainer"
cp -R /path/to/endor-labs-agent-kit/codex/endor-troubleshooter \
  "$HOME/.agents/skills/endor-troubleshooter"
cp -R /path/to/endor-labs-agent-kit/codex/findings-browser \
  "$HOME/.agents/skills/findings-browser"
cp -R /path/to/endor-labs-agent-kit/codex/malware-response \
  "$HOME/.agents/skills/malware-response"
cp -R /path/to/endor-labs-agent-kit/codex/probe-droid \
  "$HOME/.agents/skills/probe-droid"
cp -R /path/to/endor-labs-agent-kit/codex/remediation-planner \
  "$HOME/.agents/skills/remediation-planner"
cp -R /path/to/endor-labs-agent-kit/codex/sca-remediation \
  "$HOME/.agents/skills/sca-remediation"

Then invoke it from Codex:

Use the ai-sast-triage skill to triage AI SAST findings for this repository.
Use the cicd-posture skill to assess CI/CD and supply chain posture for namespace <namespace>.
Use the dependency-decision-helper skill to help with this Endor Labs workflow.
Use the package-risk-summary skill to help with this Endor Labs workflow.
Use the repository-dependency-reviewer skill to help with this Endor Labs workflow.
Use the upgrade-impact-analysis skill to help with this Endor Labs workflow.
Use the vulnerability-explainer skill to help with this Endor Labs workflow.
Use the endor-troubleshooter skill to diagnose this Endor issue from redacted error text and read-only tenant evidence.
Use the findings-browser skill to show the critical and high reachable findings for namespace <namespace>.
Use the malware-response skill to help with this Endor Labs workflow.
Use the probe-droid skill to probe GitHub org <org> for Endor monitored-branch onboarding gaps and setup prescriptions.
Use the remediation-planner skill to help with this Endor Labs workflow.
Use the sca-remediation skill to check this repository for P0 SCA findings I can start remediating.

Cursor SDK

Run generated Cursor SDK agents from Python against a local workspace or Cursor cloud repo:

python3 -m pip install -r cursor-sdk/requirements.txt
export CURSOR_API_KEY="crsr_..."
python cursor-sdk/run_cursor_agent.py endor-probe-droid-agent --workspace /path/to/repo \
  "Explain what evidence you need to assess GitHub onboarding gaps. Keep it read-only."

For cloud agents, pass --mode cloud --repo-url <repo-url> --ref <branch>.

Configure Endor Access

Access path	Used by	Notes
Endor MCP	Agents whose generated artifact declares an MCP server	Configure it through the target host's MCP mechanism only when the selected agent requires it.
endorctl api or direct Endor API	Agents that need tenant, project, finding, or policy data without MCP	The generated prompts constrain commands to documented lookups and writes. Agent or edition README files link to endorctl-setup.md when needed.
GitHub read-only inventory credentials	Probe Droid	Required when the agent compares GitHub.com repository inventory with Endor projects without cloning or mutating repositories.
Git and source-provider credentials	Mutating Claude Code agents such as AI SAST Triage and SCA Remediation	Required when the agent is expected to apply patches, open change requests, read PR/MR approval evidence, or post PR/MR comments.
Codex terminal and file-editing tools	Codex skills for mutating agents such as AI SAST Triage and SCA Remediation	The skill keeps file edits, branch pushes, PR/MR creation, PR/MR comments, and Endor policy writes behind separate approval gates.
Cursor SDK API key	Cursor SDK automation package	Required for local or cloud SDK runs. Use CURSOR_API_KEY; do not paste the key into prompts.
Endor policy-write access	AI SAST Triage standalone exceptions	Required only when a verified AppSec PR/MR approval should create a scoped Endor exception policy. The agent must show the policy spec and ask for confirmation before writing.

Example Prompts

AI SAST Triage:

@agent-ai-sast-triage triage AI SAST findings for this repository

CI/CD And Supply Chain Posture:

@agent-cicd-posture assess CI/CD and supply chain posture for namespace <namespace>

Dependency Decision Helper:

@agent-dependency-decision-helper assess npm lodash version 4.17.20

Endor Labs Package Risk Summary:

@agent-package-risk-summary summarize npm lodash version 4.17.20

Endor Labs Repository Dependency Reviewer:

@agent-repository-dependency-reviewer review this repository's dependency manifests

Endor Labs Upgrade Impact Analysis:

@agent-upgrade-impact-analysis show the safest upgrade path for repository <owner>/<repo> package lodash

Endor Labs Vulnerability Explainer:

@agent-vulnerability-explainer explain CVE-2021-44228

Endor Troubleshooter:

@agent-endor-troubleshooter diagnose this Endor scan failure from redacted error text and read-only tenant evidence

Findings Browser:

@agent-findings-browser show the critical and high reachable findings for namespace <namespace>

Malware Response Agent:

@agent-malware-response help

Probe Droid:

@agent-probe-droid probe GitHub org <org> for Endor monitored-branch onboarding gaps and setup prescriptions

Remediation Planner:

@agent-remediation-planner preview remediation options for this repository

SCA Remediation:

@agent-sca-remediation check this repository for P0 SCA findings I can start remediating

Other non-breaking low-risk UIA-backed PRs:

@agent-sca-remediation show me the other non-breaking low-risk UIA-backed PRs for this repository. Keep this separate from the P0/exploited queue and the risky solver. Do not edit files, create branches, push, or open a PR/MR.

SCA remediation PR plan:

@agent-sca-remediation prepare the top UIA-backed dependency remediation for this repository. Show the selected package, affected manifests, VersionUpgrade/UIA UUID, risk, CIA status, risk_decision, findings fixed, folded advisory/finding list, validation command, branch name, PR/MR title, and body before changing files.

🧪 Output Contract

Agents return concise prose plus a JSON block. The exact schema depends on the agent. If a signal is unavailable because of setup, authentication, account tier, or tooling, agents record that in data_gaps instead of inventing evidence.

SCA remediation outputs can be checked mechanically before a workflow advances:

endor-agent-kit validate-sca-output sca-output.json --gate selection-plan
endor-agent-kit render-sca-pr-body sca-output.json > pr-body.md
endor-agent-kit lint-sca-pr-body pr-body.md
endor-agent-kit check-install --agent sca-remediation --repo /path/to/repo
endor-agent-kit check-install --host claude-managed-agents --agent probe-droid
endor-agent-kit check-install --host codex --agent sca-remediation --skills-home ~/.agents/skills
endor-agent-kit check-install --host portable --agent sca-remediation --portable-dir /path/to/runtime/agents/sca-remediation
endor-agent-kit validate-ai-sast-output ai-sast-output.json --gate remediation
endor-agent-kit render-ai-sast-pr-body ai-sast-output.json > pr-body.md
endor-agent-kit lint-ai-sast-pr-body pr-body.md
endor-agent-kit render-ai-sast-approval-comment ai-sast-output.json > approval-comment.md
endor-agent-kit lint-ai-sast-approval-comment approval-comment.md
endor-agent-kit render-ai-sast-exception-policy-comment ai-sast-output.json > policy-comment.md
endor-agent-kit lint-ai-sast-exception-policy-comment policy-comment.md

validate-sca-output rejects Selection / Plan responses that omit risk_decision.status, use nonstandard branch names, or try to advance a CIA-indeterminate upgrade without source-usage evidence and validation requirements. render-sca-pr-body turns normalized advisory data into the AURI-style PR/MR body, including the folded advisory list, CVE-visible links to GHSA URLs, and severity emoji suffixes. check-install catches copied Claude Code primary artifacts, staged Claude Managed Agents bundles, and installed Codex skill bundles that are stale versus the checked-in Agent Kit catalog. It can also compare a copied portable bundle with the catalog when you pass --host portable and --portable-dir.

AI SAST triage outputs can be checked before remediation, PR/MR, or exception-policy gates advance. validate-ai-sast-output requires project and namespace provenance, finding/source-location provenance, approval evidence before exception policies, and a rendered PR/MR body when a remediation change request is part of the plan. For exception policies it also checks accepted-risk expiration, approval reason matching, approved finding scope, project selector scope, policy names, idempotency checks, and human-readable decision comments. The AI SAST PR/MR renderer follows the AURI AI SAST remediation structure with auri:ai-sast-context metadata, severity indicator emojis, sanitized AURI evidence, a standalone exception-request prompt block, folded finding details, and standalone Agent Kit policy-write gates that still require independent AppSec approval before any Endor policy write. Standalone Agent Kit is not a webhook listener: PR/MR comments are approval evidence, and a user or external automation must invoke the installed agent before any policy can be created or reused.

For sca-remediation, keep the three remediation lanes distinct:

• P0/exploited remediation candidates: rank reachable or exploited critical/high findings and require UIA evidence before naming a best fix.
• Other non-breaking low-risk UIA-backed PRs: list only low-risk, CIA-clean recommendations with zero introduced findings and enough repo metadata to open a PR/MR.
• Risky or indeterminate upgrades: use risk_decision from Endor evidence plus local source usage before applying or opening a change request.

The low-risk lane reports low_risk_recommendations, candidate_prs, ready_to_open, most_findings_in_one_pr, and p0_duplicates_hidden. Validation commands must come from the repository's actual manifest and package-manager layout; the agent must not assume Java, Maven, npm, Python, Go, .NET, Ruby, Rust, or any other ecosystem from prior runs.

🛡️ Safety Model

Most agents in this kit are read-only. Recipes declare their safety class and host capabilities explicitly.

Read-only agents do not:

• edit files
• create pull requests
• run scans
• dismiss findings
• create policies
• mutate Endor Labs state

Mutating agents are published only when their recipe declares the required host capabilities. AI SAST Triage and SCA Remediation may fetch source context, write patch files, run git/source-provider commands, and open a change request when the user asks for that workflow and the target repository credentials are available.

When a read-only agent permits Bash, its prompt limits Bash to documented read-only Endor lookup commands. Claude Code artifacts deny Bash when it is not needed.

When a recipe declares host_capabilities_required.read_files: true, Claude Code artifacts allow only Read, Glob, Grep, and LS for read-only workspace inspection; file mutation, notebook, web, and todo tools remain denied. Claude Managed Agents artifacts omit the pre-built agent toolset unless an agent needs read-only Bash, and then enable only Bash with confirmation.

Guardrail And Runtime Security Docs

The repository includes two maintainer-facing guardrail references:

• docs/guardrails.md maps recipe, host, portable, workflow, and artifact-integrity controls to the current catalog.
• docs/portable-runtime-conformance.md defines the runtime controls a customer-managed portable integration must enforce around adapters, approvals, evidence, audit, untrusted content, and failure modes.

Contribute An Agent

This repository is the source of truth for agent behavior and generated package output. The public endorlabs/ai-plugins repository is the distribution mirror. New agents, skills, hooks, action contracts, and publication changes start here, not in ai-plugins.

Read docs/contributing-agents.md for the official proposal, review, approval, provenance, signing, and generated ai-plugins PR process.

Create Agents With The Skill

Use the Create Endor Labs Agent skill to make your own Endor Labs agent. The skill lives at skills/create-endor-labs-agent/SKILL.md and guides an assistant through agent design, recipe authoring, prompt sections, evals, architecture diagrams, tests, catalog regeneration, and validation.

The skill supports two public input styles: a net-new agent brief or a generic sanitized agent blueprint. Private tools may generate a blueprint, but Agent Kit should only receive customer-safe recipe, action, instruction, eval, and architecture source files.

For Claude Code, install it at either the repository or user level:

# Repository-level install
mkdir -p .claude/skills
cp -R skills/create-endor-labs-agent .claude/skills/

# User-level install
mkdir -p ~/.claude/skills
cp -R skills/create-endor-labs-agent ~/.claude/skills/

Then ask your assistant:

Use the create Endor Labs agent skill to make an agent that <does the workflow you want>.

You can also point any agent directly at skills/create-endor-labs-agent/SKILL.md if it does not support native skills.

Development Setup

python3 -m venv .venv
. .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e ".[dev]"

Authoring Workflow

1. Edit source/agents/<agent>/recipe.yaml.
2. Edit source/agents/<agent>/instructions.md.
3. Update source/agents/<agent>/evals/cases.yaml.
4. Add source/agents/<agent>/architecture.svg in the existing diagram format.
5. Add or update tests under tests/.
6. Validate and regenerate the catalog.

endor-agent-kit validate source/agents/<agent>/recipe.yaml
endor-agent-kit doctor-new-agent source/agents/<agent>/recipe.yaml
endor-agent-kit authoring-check source/agents/<agent>/recipe.yaml --new-agent
endor-agent-kit publish source/agents/*/recipe.yaml --dest . --prune --include-plugins
endor-agent-kit check-guardrails --catalog-root .
endor-agent-kit verify-provenance --catalog-root .
endor-agent-kit verify-endor-context --upstream
python -m pytest -q
git diff --exit-code -- README.md manifest.json .agents/plugins .claude-plugin .cursor-plugin agents assets claude-code claude-managed-agents codex cursor-sdk gemini hooks plugins portable skills

Pull requests should include both source changes and regenerated artifacts. CI runs validation, generated-artifact drift checks, guardrails, and provenance verification. After a maintainer merges to main, .github/workflows/publish-ai-plugins-pr.yml opens or updates the generated ai-plugins distribution PR.

CLI Reference

Command	Purpose
endor-agent-kit validate source/agents/<agent>/recipe.yaml	Validate one recipe.
endor-agent-kit doctor-new-agent source/agents/<agent>/recipe.yaml	Run contributor-facing pre-PR checks and next-step guidance for a new source agent.
endor-agent-kit authoring-check source/agents/<agent>/recipe.yaml --new-agent	Check source-first authoring rules for a new public agent.
endor-agent-kit compile source/agents/<agent>/recipe.yaml --target <host>	Compile one recipe into its local dist/ directory.
endor-agent-kit compile source/agents/<agent>/recipe.yaml --target <host> --edition <edition>	Compile one edition for one host.
endor-agent-kit publish source/agents/*/recipe.yaml --dest . --prune	Regenerate the checked-in catalog and remove stale generated agents.
endor-agent-kit check-guardrails --catalog-root .	Check generated catalog artifacts against guardrail policies.
endor-agent-kit verify-provenance --catalog-root .	Verify generated catalog artifacts against recorded manifest checksums.
endor-agent-kit verify-endor-context --upstream	Compare committed Endor OpenAPI/docs provenance with live upstream sources.
endor-agent-kit refresh-endor-context	Refresh source/endor-context/provenance.json after intentional upstream Endor API/docs changes.
endor-agent-kit validate-sca-output sca-output.json --gate selection-plan	Validate structured sca-remediation output before advancing a workflow gate.
endor-agent-kit render-sca-pr-body sca-output.json > pr-body.md	Render the AURI-style SCA remediation PR/MR body from normalized JSON.
endor-agent-kit lint-sca-pr-body pr-body.md	Lint a rendered SCA remediation PR/MR body for required sections, advisory formatting, and severity suffixes.
endor-agent-kit validate-ai-sast-output ai-sast-output.json --gate remediation	Validate structured ai-sast-triage output before remediation, PR/MR, or exception gates advance.
endor-agent-kit render-ai-sast-pr-body ai-sast-output.json > pr-body.md	Render an AURI-style AI SAST remediation PR/MR body from normalized JSON.
endor-agent-kit lint-ai-sast-pr-body pr-body.md	Lint an AURI-style AI SAST remediation PR/MR body for required sections, hidden context metadata, and severity indicators.
endor-agent-kit render-ai-sast-approval-comment ai-sast-output.json > approval-comment.md	Render a standalone AppSec approval request comment.
endor-agent-kit lint-ai-sast-approval-comment approval-comment.md	Lint the approval request comment and exact approval phrase.
endor-agent-kit render-ai-sast-exception-policy-comment ai-sast-output.json > policy-comment.md	Render a human-readable Endor exception policy decision comment.
endor-agent-kit lint-ai-sast-exception-policy-comment policy-comment.md	Lint the policy decision comment for policy name/UUID, project label, evidence, and raw selector leakage.
endor-agent-kit check-install --agent sca-remediation --repo /path/to/repo	Check whether a copied repo-level Claude Code agent matches the generated catalog artifact.
endor-agent-kit check-install --host claude-managed-agents --agent probe-droid	Check whether a staged Claude Managed Agents bundle matches the generated catalog bundle.
endor-agent-kit check-install --host codex --agent sca-remediation --skills-home ~/.agents/skills	Check whether an installed Codex skill directory matches the generated catalog bundle.
endor-agent-kit check-install --host portable --agent sca-remediation --portable-dir /path/to/runtime/agents/sca-remediation	Check whether a copied portable bundle matches the generated catalog bundle.

Supported compile targets are claude-code, claude-managed-agents, codex, gemini, portable, and raw.

📖 Reference

Recipe Reference

Recipes are YAML files with schema version 1 or 2. They describe the agent's prompt, Endor access paths, host capabilities, inputs, outputs, evals, and published host editions. Schema v2 recipes may also point to actions.yaml for semantic side-effect contracts such as opening change requests or requesting exception-policy approval.

Field	Purpose
id, name, version, description	Public catalog identity and copy.
safety_class, mutations	Safety contract. Recipes may be read_only, dry_run, or explicitly mutating with matching host capabilities.
supported_transports	Endor access paths such as mcp and endorctl_api.
host_capabilities_required	Abstract host capabilities that compilers map to host-specific tools.
action_contracts_path	Optional schema v2 path to actions.yaml, which declares semantic side effects and adapter requirements.
inputs, outputs	User-facing IO contract and expected JSON output shape.
compatible_hosts	Hosts that should receive generated artifacts.
host_editions	Optional host-specific edition selection. Omit to publish all default editions for that host.
required_endor_mcp_tools, endorctl_api_invocations	Endor tools and API lookup groups the prompt may use.
instructions_path, evals	Source prompt and eval case files relative to the recipe.
architecture.svg	Required source diagram copied into generated catalog artifacts when present.

Generated artifacts must not be edited as the first step. Change the recipe or instructions source, publish the catalog, then review the generated diff.

Repository Reference

Layout

source/
  agents/
    <agent>/
      recipe.yaml
      actions.yaml
      instructions.md
      evals/cases.yaml
agents/
  <generated-cursor-agent>.md
skills/
  create-endor-labs-agent/
    SKILL.md
  <generated-cursor-skill>/
    SKILL.md
    architecture.svg
.cursor-plugin/
  marketplace.json
  plugin.json
cursor-sdk/
  README.md
  agent_definitions.json
  run_cursor_agent.py
  requirements.txt
  agents/
    <generated-cursor-sdk-agent>.md
assets/
  logo.png
hooks/
  hooks.json
  *.sh
docs/
  distribution-sync.md
  for-agents.md
  getting-started.md
  plugin-packaging-design.md
  plugin-release-checklist.md
  maintainer-guide.md
src/endor_agent_kit/
tests/
plugins/
  claude/endor-labs-agent-kit/
  codex/endor-labs-agent-kit/
  gemini/endor-labs-agent-kit/
claude-code/<agent>/
claude-managed-agents/<agent>/
codex/<agent>/
gemini/<agent>/
portable/<agent>/
manifest.json

Manifest

manifest.json lists published artifacts and their SHA-256 checksums. Each manifest entry also records source.builder_recipe, which points back to the recipe that generated the artifact set.

Generated Catalog

The root catalog directories are intentionally checked in:

• claude-code/
• claude-managed-agents/
• codex/
• gemini/
• portable/
• plugins/
• .cursor-plugin/
• cursor-sdk/
• agents/<generated-cursor-agent>.md
• skills/<generated-cursor-skill>/
• hooks/
• assets/logo.png
• manifest.json

These paths are customer-facing and should stay stable.

Release And License

Before publishing a public release, verify that generated artifacts are fresh and that the MIT license text is present in LICENSE and package metadata.

Agent Kit is published under the MIT license.