prefactor-skill-selector

Prefactor skill selector

Use this file as a router for Prefactor skills.

Available skills

• skills/bootstrap-existing-agent-with-prefactor-cli/SKILL.md: provision Prefactor credentials with prefactor setup / prefactor setup --create, write env values, and choose a package.
• skills/instrument-existing-agent-with-prefactor-sdk/SKILL.md: instrument an existing agent with Prefactor SDK so coding tools can see runs, llm/tool spans, token usage, and failures.
• skills/create-provider-package-with-core/SKILL.md: create a new provider package as a thin adapter over @prefactor/core with core-first boundaries and tracing conventions.
• skills/report-agent-risk-data/SKILL.md: populate data_risk fields on span types for compliance tracking and data governance after an agent is already instrumented.
• skills/write-span-summary-templates/SKILL.md: write Liquid display templates on span type schemas so traces show readable one-line summaries in the Prefactor UI.
• skills/inspect-agent-run-with-prefactor-cli/SKILL.md: perform root-cause analysis on a Prefactor agent run from CLI span data; run in the agent's own codebase (user provides instance ID and symptom).

Selection rules

• If the request is about provisioning Prefactor resources or credentials via CLI, load skills/bootstrap-existing-agent-with-prefactor-cli/SKILL.md first.
• If the request is about adding telemetry to an existing agent without rewriting business logic, load skills/instrument-existing-agent-with-prefactor-sdk/SKILL.md.
• If the request is about creating a custom provider adapter with @prefactor/core, load skills/create-provider-package-with-core/SKILL.md.
• If the request is about adding risk or compliance metadata to span types, load skills/report-agent-risk-data/SKILL.md.
• If the request is about span summary templates, display templates, Liquid templates on schemas, or making spans readable in the Prefactor UI, load skills/write-span-summary-templates/SKILL.md.
• If the request is about root-cause analysis on a bad, surprising, expensive, incomplete, or downvoted agent run, load skills/inspect-agent-run-with-prefactor-cli/SKILL.md.

Default workflow

When instrumenting an existing agent, default to this order:

1. Confirm the human has run prefactor login.
2. Run skills/bootstrap-existing-agent-with-prefactor-cli/SKILL.md (prefactor setup or prefactor setup --create --name ... --description ...).
3. Inspect the project and install the matching Prefactor adapter (@prefactor/langchain, @prefactor/ai, @prefactor/openclaw, @prefactor/claude).
4. For adapter-style instrumentation, keep init, withSpan, and shutdown imports from that same adapter package (or pass an explicit tracer when using core withSpan).
5. If no matching adapter package exists, use skills/create-provider-package-with-core/SKILL.md.
6. Instrument the existing agent with skills/instrument-existing-agent-with-prefactor-sdk/SKILL.md.
7. Optionally populate compliance metadata with skills/report-agent-risk-data/SKILL.md.
8. Optionally add readable trace summaries with skills/write-span-summary-templates/SKILL.md.