Codex Supercharge Router

Use this first when the request is broad. Pick one primary skill, then add at most one support skill.

Routing Table

User need	Primary skill	Support skill
"I am running out of context"	$low-token-context	$model-cost-router
"Understand this repo"	$repo-intelligence-brief	$codebase-graph-scout
"Make this repo easier to revisit"	$repo-to-wiki	$agent-memory-ledger
"Use or test an MCP server"	$mcp-server-lab	$mcp-safety-review
"Add MCP CI/conformance gates"	$auto-skill-build-mcp-conformance-harness	$mcp-safety-review
"Is this MCP safe?"	$mcp-safety-review	$sandbox-source-intake
"Clean up MCP config secrets"	$auto-skill-safety-mcp-secret-guard	$mcp-safety-review
"Turn an API into tools"	$openapi-to-mcp-skill	$mcp-safety-review
"Build, redesign, or review frontend UI"	$open-design-ui-library	$low-token-context
"Ship a reliable agent workflow"	$auto-skill-build-agent-reliability-loop	$agent-observability
"Track local agent runs/costs"	$agent-observability	$model-cost-router
"Append or review recent run history"	$auto-skill-build-agent-run-ledger-dashboard	$agent-observability
"Grade Supercharge edits or share status"	$auto-skill-build-supercharge-feedback-loop	$agent-observability
"Run bounded iterative experiments"	$experiment-loop-runner	$agent-memory-ledger
"Choose model tier or cost route"	$model-cost-router	$low-token-context
"Import third-party skill/prompt pack"	$auto-skill-safety-external-skill-intake	$skill-quality-gate
"Create reusable prompt templates"	$codex-prompt-template-builder	$skill-quality-gate
"Create many useful skills"	$skill-pack-factory	$skill-quality-gate
"Audit my skill/plugin pack"	$skill-quality-gate	$low-token-context
"Clone fresh repos safely"	$sandbox-source-intake	$repo-intelligence-brief
"Synthesize patterns across cloned repos"	$auto-skill-build-source-pattern-synthesis	$skill-pack-factory
"Verify GitHub trend or discovery claims"	$auto-skill-trend-github-discovery-radar	$sandbox-source-intake
"Set up a day/week workflow"	$workflow-launchpad	$agent-memory-ledger
"Set up daemon/status/worker control"	$workflow-launchpad	$mcp-safety-review

Boundary Rules

• If the request affects live users, external tools, accounts, data, payments, or production budgets, choose the reliability or safety skill before local observability.
• If the user asks for CI, release gates, JUnit/XML, conformance, or tool-surface diffs, choose $auto-skill-build-mcp-conformance-harness over $mcp-server-lab.
• If the task is one bounded trial with a stop condition, use $experiment-loop-runner; if it feeds production guardrails or canaries, use $auto-skill-build-agent-reliability-loop.
• If the task is only model choice or token/cost measurement, use $model-cost-router; if routing is enforced by gateway budgets or fallback policy, use $auto-skill-build-agent-reliability-loop.
• If the source is an external skill pack, slash-command pack, hook plugin, or prompt catalogue, run $auto-skill-safety-external-skill-intake before $skill-pack-factory.
• If the task creates or adapts reusable Codex prompt templates, use $codex-prompt-template-builder. If the template pack came from an external source, run $auto-skill-safety-external-skill-intake first; run $skill-quality-gate before publishing.
• If the task compares multiple local source clones before deciding what to build, use $auto-skill-build-source-pattern-synthesis before $skill-pack-factory.
• If the task uses current popularity, trend, stars, topics, "latest", or public launch/discoverability claims, use $auto-skill-trend-github-discovery-radar before writing README, marketplace, or skill-generation conclusions.
• If the task has daemons, workers, queues, sidecars, task APIs, status/log commands, or multi-agent leases, use $workflow-launchpad before starting processes. Add $mcp-safety-review only when the control surface exposes side-effecting tools or credentials.
• If the task creates, edits, reviews, grades, or validates Codex Supercharge artifacts, append a $auto-skill-build-supercharge-feedback-loop batch so future sessions and automations can see the latest state.
• If the task turns an OpenAPI/Swagger spec into tools, use $openapi-to-mcp-skill for the surface plan. Add $mcp-safety-review before serving or installing any generated bridge when credentials, writes, destructive operations, admin actions, or account data are involved.
• If the task is visual/frontend work, use $open-design-ui-library for design-system selection and craft checks. Use $workflow-launchpad only when process control, local daemons, or worker lifecycle is the main risk.

Skip When

• The user has explicitly named the exact skill or plugin to use.
• The task is a narrow code edit with an obvious local workflow.

Default Policy

• Prefer small, composable workflows over large magic systems.
• Prefer local markdown state over repeated long prompts.
• Prefer shallow clones and README/manifest inspection before installs.
• Prefer source evidence over trend-list claims.
• When a repo is unverified or risky, classify it as inspiration only.

Read references/source-repos.md when you need the cloned source inventory. Read references/source-pattern-synthesis.md when comparing source patterns. Read references/github-discovery-radar.md when claims depend on current GitHub evidence, public repo status, stars, topics, or trend positioning. Read references/workflow-control-plane.md when launching daemon, worker, status/log, task API, or multi-agent coordination workflows. Read references/supercharge-feedback-loop.md before continuing prior Codex Supercharge artifact work or handing status to subagents/automations. Read references/openapi-tool-surface.md when converting API specs into MCP or agent tool surfaces. Read references/open-design-ui-library.md when selecting UI systems, tokens, or craft checks. Read references/prompt-template-system.md and prompt-templates/README.md when creating or adapting reusable prompt templates. Read references/skill-routing-boundaries.md when routing overlaps.

Validation

• Choose one primary skill and at most one support skill.
• Escalate safety/reliability before execution when accounts, credentials, live users, or destructive tools are involved.
• If routing overlaps, cite the boundary rule used.