agent-selector

L1 Improvement

• Reframed the selector playbook to follow Skill Forge required sections with explicit routing notes and completion checks.
• Added prompt-architect style constraint capture, confidence ceilings, and memory tagging to keep selections auditable.

---

LIBRARY-FIRST PROTOCOL (MANDATORY)

Before writing ANY code, you MUST check:

Step 1: Library Catalog

• Location: .claude/library/catalog.json
• If match >70%: REUSE or ADAPT

Step 2: Patterns Guide

• Location: .claude/docs/inventories/LIBRARY-PATTERNS-GUIDE.md
• If pattern exists: FOLLOW documented approach

Step 3: Existing Projects

• Location: D:\Projects\*
• If found: EXTRACT and adapt

Decision Matrix

Match	Action
Library >90%	REUSE directly
Library 70-90%	ADAPT minimally
Pattern exists	FOLLOW pattern
In project	EXTRACT
No match	BUILD (add to library after)

---

STANDARD OPERATING PROCEDURE

Purpose

Route work to the highest-quality specialist agent by analyzing task intent, constraints, tools, and historical performance.

Trigger Conditions

• Positive: choosing an agent for a task, validating an existing agent assignment, planning multi-agent workflows.
• Negative/reroute: creating a new agent (agent-creator/agent-creation) or optimizing prompts without agent routing (prompt-architect).

Guardrails

• Require English outputs with explicit confidence ceilings for the selection rationale.
• Never return a generic agent when a specialized option exists; surface top alternatives with scores.
• Avoid opaque reasoning—show matched capabilities, tools, and decision factors.
• Respect registry integrity: only return agents that exist and are permitted for the task.

Execution Phases

1. Intent & Constraint Capture: Extract domain, phase, tools, risk level, and success criteria; mark hard vs soft constraints.
2. Candidate Retrieval: Query registry/memory for agents; narrow by category, tools, and phase.
3. Scoring & Ranking: Score by capability match, domain fit, tool alignment, and historical outcomes; include tie-break rules.
4. Recommendation & Rationale: Provide selected agent, alternatives, reasoning, and integration notes; state confidence ceiling.
5. Validation Loop: If no suitable agent, recommend creation path and log gaps for skill-forge backlog.

Pattern Recognition

• Tool-constrained tasks → prioritize agents with matching tool whitelists and error-handling experience.
• High-risk domains (security/compliance) → bias toward agents with safety guardrails and audits.
• Multi-phase projects → pick agents aligned to current phase (planning vs build vs test).

Advanced Techniques

• Use semantic + rules-based scoring to avoid tool-only bias.
• Apply cooldowns for recently failing agents and boost for proven recent runs.
• Capture reason codes for why top alternatives were not chosen.

Common Anti-Patterns

• Selecting based solely on name similarity.
• Ignoring tool requirements or phase context.
• Omitting confidence ceilings or rationale in the output.

Practical Guidelines

• Provide 1 primary and 2-3 backup agents with concise reasons and score deltas.
• Include file paths for reproducibility and auditing.
• Encourage small validation task before full delegation when confidence <0.85.

Cross-Skill Coordination

• Upstream: prompt-architect for refined task descriptions; skill-forge for complete agent metadata.
• Downstream: delivery/ops skills consuming the selected agent; recursive-improvement when selection fails.

MCP Requirements

• Memory/vector search recommended for registry lookup; tag WHO=agent-selector-{session}, WHY=skill-execution.
• Ensure access controls before suggesting agents tied to sensitive tools.

Input/Output Contracts

inputs:
  task: string  # required description of the work
  domain: string  # optional domain hint
  phase: string  # optional project phase
  tools: list[string]  # optional required tools/MCP servers
outputs:
  primary_agent: object  # name, path, score, capabilities, tools
  alternatives: list[object]  # ranked backups with reasons
  rationale: string  # key decision factors and constraints satisfied

Recursive Improvement

• Feed mis-selections and user feedback into recursive-improvement to tune scoring weights and filters.

Examples

• Select a backend API agent for an Express+PostgreSQL feature with OpenAPI contract checks.
• Recommend a security review agent for dependency audit with SBOM parsing and CVE triage.

Troubleshooting

• No match found → relax soft constraints, widen category, or trigger agent creation.
• Multiple top ties → expose tie-break factors (tools, recency, safety posture).
• Agent returned lacks required tool → fail selection and propose alternatives with matching tools.

Completion Verification

• Primary agent and backups returned with scores and reasoning.
• Constraints and tools addressed; registry paths included.
• Confidence ceiling stated; memory tags applied if MCP used.
• Escalation path defined when no suitable agent exists.

Confidence: 0.70 (ceiling: inference 0.70) - Selector SOP aligned to Skill Forge structure with prompt-architect ceilings.