Provide a compact human-readable routing and navigation layer for AgentDefaults while agentdefaults.manifest.json remains the authoritative machine-readable featured-stack registry.
| Need | Start with |
|---|---|
| Operate the bounded loop quickly | docs/loops/QUICK_REFERENCE.md |
| Understand the full bounded lead/reviewer loop | docs/loops/README.md |
| Run a bounded lead/reviewer completion loop in VS Code Copilot | docs/quickstarts/bounded-completion.md |
| Choose among the principal and specialist engineering agents | ENGINEERING_AGENTS_INDEX.md |
| Engineer or harden GitHub Actions workflows/actions | docs/quickstarts/github-actions-engineer.md |
| Use authenticated Comet browser research | docs/quickstarts/comet-authenticated-research.md |
| Apply the Token Economy stack | docs/quickstarts/token-economy.md |
| Use OpenAI Codex | docs/quickstarts/codex.md |
| Use Claude Code | docs/quickstarts/claude.md |
| Edit local YouTube footage with Claude Code | docs/quickstarts/claude-video-editing.md |
| Understand Claude project hooks / Graft adapter | .claude/README.md |
| Use GitHub Copilot custom agents | .github/agents/ |
| Understand Copilot prompt adapters | .github/prompts/README.md |
| Use Gemini / Gemini CLI | GEMINI.md |
| Use a generic repository-aware agent | AGENTS.md |
| Adapt examples safely | examples/README.md |
| Understand loop configuration | config/README.md |
| See all tool mappings | docs/tool-integration-guide.md |
| Run the Humanizer pass on human-facing prose | skills/humanizer.md |
| Decide when to use an external tool (Hyperframes, Apify MCP, OpenSEO, Agent Skills, Codeburn, Ponytail) | docs/external-tools.md |
Add the pointer block to an app repo's AGENTS.md |
docs/snippets/downstream-agents-pointer.md |
| Validate the repository | scripts/validate-agentdefaults.py |
Use docs/quickstarts/bounded-completion.md when implementation/qualification work needs a single Integration Owner, independent adversarial review, durable loop state, deterministic verification, bounded continuation, visual-artifact evidence, and an objective completion gate.
For active operation, keep docs/loops/QUICK_REFERENCE.md open. For ownership, state, stale-evidence, approval, Stop-hook, recovery, and loop-design semantics, use the full docs/loops/README.md.
agents/bounded-completion-lead.md
agents/bounded-completion-reviewer.md
skills/bounded-completion-orchestration.md
schemas/bounded-completion-task.schema.json
config/bounded-completion.json
scripts/bounded-completion.py
.github/agents/bounded-completion-lead.agent.md
.github/agents/bounded-completion-reviewer.agent.md
The custom agents intentionally omit model: bindings until exact qualified local model identifiers are repository-discoverable. The operator guide documents manual VS Code model-picker selection for the intended Qwen lead/reviewer roles.
| Scope | Canonical agent | Skill | Quickstart | Copilot adapter |
|---|---|---|---|---|
| DevOps/platform | agents/principal-devops-engineer.md |
skills/production-devops-engineering.md |
docs/quickstarts/principal-devops-engineer.md |
.github/agents/principal-devops-engineer.agent.md |
| AI engineering | agents/principal-ai-engineer.md |
skills/production-ai-engineering.md |
docs/quickstarts/principal-ai-engineer.md |
.github/agents/principal-ai-engineer.agent.md |
| Materially cross-domain AI + DevOps | agents/principal-ai-devops-engineer.md |
skills/production-ai-devops-engineering.md |
docs/quickstarts/principal-ai-devops-engineer.md |
.github/agents/principal-ai-devops-engineer.agent.md |
Use the smallest correct owner. Infrastructure hosting an AI workload is still DevOps-only unless AI application behavior also requires material change. GitHub Actions-specific workflow/runtime work should use the specialist route below before the generic DevOps route.
| Need | Start with |
|---|---|
GitHub Actions workflow/action engineering, reusable workflows, runner trust, GITHUB_TOKEN, OIDC, Dependabot/fork boundaries, Actions release automation, artifacts/caches, concurrency/reruns, or qualification |
docs/quickstarts/github-actions-engineer.md · agents/github-actions-engineer.md |
Work specifically on Quazmoz/K8SHomelab Kubernetes/Flux GitOps, app deployment, storage/networking, or cluster incidents |
docs/quickstarts/kubernetes-homelab-engineer.md · agents/kubernetes-homelab-engineer.md |
| Audit or harden DevOps cybersecurity across Terraform/OpenTofu, Ansible/AAP, Jenkins, CI/CD, GitOps, IAM, secrets, runners/agents, state, or software supply chain | docs/quickstarts/devsecops-security-engineer.md · agents/devsecops-security-engineer.md |
| Create, audit, or reconcile DevOps docs-as-code, runbooks, Markdown, Mermaid, and documentation diagrams | docs/quickstarts/devops-documentation-engineer.md · agents/devops-documentation-engineer.md |
| De-slop an existing codebase: stale comments/docstrings, duplicate helpers, dead residue, abstraction inflation, brittle tests, dependency/config drift, weak failure handling, or practical efficiency cleanup across languages | docs/quickstarts/codebase-maintenance-engineer.md · agents/codebase-maintenance-engineer.md |
| Automate Google Play releases, RevenueCat monetization configuration, AdMob inventory, or portfolio-wide local build -> Internal Testing orchestration | docs/quickstarts/mobile-release-automation.md · agents/mobile-release-automation-engineer.md |
| Design, build, or audit another AI agent | docs/quickstarts/agent-builder.md |
| Choose or challenge an automation platform | AUTOMATION_PLATFORM_INDEX.md |
docs/quickstarts/github-actions-engineer.md
agents/github-actions-engineer.md
skills/github-actions-engineering.md
prompts/implementation/github-actions-task.md
schemas/github-actions-task.schema.json
examples/github-actions-task.yaml
docs/github-actions-engineer-acceptance-tests.md
.github/agents/github-actions-engineer.agent.md
scripts/validate-github-actions-stack.py
This specialist owns GitHub Actions-specific event trust, fork/Dependabot behavior, pull_request_target/privileged workflow_run risks, token/secret/OIDC scope, reusable-workflow contracts, action/workflow pinning and provenance, GitHub-hosted/self-hosted runner boundaries, caches/artifacts, concurrency/cancellation/reruns, release/deployment automation, artifact identity, and Actions cost controls.
Use prompts/implementation/github-actions-task.md for repeatable work and schemas/github-actions-task.schema.json when an orchestrator needs a machine-verifiable authority/trust/verification contract.
docs/quickstarts/mobile-release-automation.md
agents/mobile-release-automation-engineer.md
skills/mobile-release-automation-orchestration.md
skills/android-portfolio-release-orchestration.md
skills/google-play-release-automation.md
skills/google-play-localized-pricing.md
skills/revenuecat-monetization-automation.md
skills/admob-inventory-automation.md
prompts/implementation/mobile-release-automation-task.md
prompts/implementation/android-portfolio-release-orchestration.md
prompts/implementation/google-play-localized-pricing-rollout.md
schemas/mobile-release-automation-task.schema.json
examples/mobile-release-automation-task.yaml
docs/mobile-release-automation-acceptance-tests.md
tools/mobile-release-automation/
.github/agents/mobile-release-automation-engineer.agent.md
scripts/validate-mobile-release-automation-stack.py
This specialist owns Play app bundle upload, testing tracks, staged rollout and promotion, Play monetization products, RevenueCat apps/products/entitlements/offerings/packages, and AdMob apps and ad units. For many local Android/Wear repositories, skills/android-portfolio-release-orchestration.md adds safe checkout discovery, per-repo Gradle qualification, exact-candidate manifests, and verified Internal Testing uploads without making the portfolio registry authoritative for build state.
The three platforms are not equally automatable, and the stack treats that as a first-class fact. Google Play and RevenueCat support API-driven creation. AdMob accepts OAuth user credentials only, rejects service accounts entirely, and gates app and ad-unit creation per account through a Google account manager. Probe AdMob access before planning AdMob automation.
Use prompts/implementation/google-play-localized-pricing-rollout.md for approval-gated portfolio pricing work: it starts in read-only planning mode, produces per-app immutable plan IDs, stops for explicit approval, and only then permits exact-plan application plus read-back verification.
tools/mobile-release-automation/ is the runnable local toolkit: a CLI, an offline test suite, and a local stdio MCP server for Play and AdMob, for which no first-party MCP server exists. Use RevenueCat's first-party server at https://mcp.revenuecat.ai/mcp for RevenueCat.
| Need | Start with |
|---|---|
| Google Play growth / ASO / listing localization review | docs/quickstarts/google-play-growth.md |
| Palmier Pro MCP video editing | docs/quickstarts/palmierpro-mcp.md |
| Claude Code local video editing | docs/quickstarts/claude-video-editing.md |
| App-market browser research | docs/quickstarts/app-market-research.md |
| Community app-idea validation | docs/quickstarts/community-app-validation.md |
| Token economy | docs/quickstarts/token-economy.md |
| Wear OS development | WEAROS_DEVELOPMENT_INDEX.md |
| Wear OS Play Store release | WEAROS_INDEX.md |
| US-Europe travel preparation | TRAVEL_INDEX.md |
Some canonical agents are intentionally not registered as featured stacks but still need explicit operating guidance.
| Need | Start with |
|---|---|
| Authenticated local Comet research with human auth handoff and prompt-injection boundaries | docs/quickstarts/comet-authenticated-research.md |
| Understand all canonical agents and which ones need quickstarts | agents/README.md |
| Understand skill families/composition | skills/README.md |
| Understand prompt categories | prompts/README.md |
The authoritative stack composition is agentdefaults.manifest.json. It currently registers:
- Agent Architect and Builder
- Codebase Maintenance and De-Slop Engineering
- Bounded Two-Agent Completion
- Principal AI and DevOps Engineering
- Principal DevOps Engineering
- GitHub Actions Engineering
- Kubernetes Homelab Engineering
- DevSecOps Security Engineering
- DevOps Documentation Engineering
- Principal AI Engineering
- Automation Platform Architecture and Selection
- Google Play Growth Optimization
- Mobile Release and Monetization Automation
- Palmier Pro MCP Video Editing
- Claude Code Local Video Editing
- App Market Browser Research
- Community App Idea Validation
- Token Economy
- Wear OS Development
- Wear OS Play Store Release
- US-Europe Travel Prep
agents/ complete reusable agent profiles
skills/ composable behavior/task modules
prompts/ reusable task and benchmark prompts
schemas/ machine-readable contracts for structured workflows
Tool-specific entrypoints and wrappers route to these files; they are not separate canonical implementations.
OpenAI Codex -> AGENTS.md
Claude Code -> CLAUDE.md -> @AGENTS.md (+ optional .claude runtime adapters)
GitHub Copilot -> .github/copilot-instructions.md + .github/agents/*.agent.md + .github/prompts/*.prompt.md
Gemini -> GEMINI.md
Generic repo-aware agent -> AGENTS.md
See docs/tool-integration-guide.md for details.
Run:
python3 scripts/validate-agentdefaults.pyThe canonical suite covers structure, schemas, manifest integrity, stack invariants, Markdown links, cross-tool entrypoints, principal engineering contracts, GitHub Actions contracts/routing, specialist DevOps documentation contracts, DevSecOps security contracts, codebase-maintenance/de-slop contracts, and bounded-completion control-plane regressions.
For GitHub Actions stack changes, the focused check is:
python3 scripts/validate-github-actions-stack.pyDo not report a validator as passed unless it actually ran successfully.