Meta Tools

L1 Improvement

• Rebuilt the meta-tools guidance with Skill Forge required sections, tool composition guardrails, and verification steps.
• Added prompt-architect style constraint extraction plus confidence ceilings for tool recommendations.

STANDARD OPERATING PROCEDURE

Purpose

Create and evolve a library of development tools with clear contracts, composition patterns, and validation evidence so downstream agents can reliably orchestrate them.

Trigger Conditions

• Positive: requests to design a new tool, improve an existing one, or compose multiple tools into a workflow.
• Negative/reroute: single-use prompts (prompt-architect), agent design (agent-creator), or pure skill scaffolding (skill-builder/skill-forge).

Guardrails

• Define explicit inputs/outputs, side effects, and failure modes for every tool.
• Include safety and rate-limit considerations for external integrations.
• Maintain English outputs with explicit confidence ceilings for recommendations.
• Require tests (unit/simulation) before promoting a tool for orchestration use.

Execution Phases

1. Discovery: Capture problem, environment, constraints, and integration surfaces; classify hard/soft/inferred constraints.
2. Interface Design: Specify input/output contracts, error semantics, and idempotency expectations.
3. Implementation/Selection: Build or adapt tools; ensure portability and minimal dependencies.
4. Validation: Run functional, boundary, and misuse tests; record evidence and metrics.
5. Composition: Document how the tool chains with others, including pre/post hooks.

Pattern Recognition

• IO-transformer tools → emphasize deterministic formats and schema validation.
• External-service tools → highlight auth, retries, and timeout strategies.
• Orchestration helpers → clarify preconditions and postconditions for safe chaining.

Advanced Techniques

• Provide composable adapters (wrappers) to normalize outputs for downstream tools/agents.
• Use circuit-breaker patterns for unstable integrations.
• Capture provenance metadata so agent-selector can trust the tool pedigree.

Common Anti-Patterns

• Ambiguous side effects or undocumented environment requirements.
• Overloaded tools that do too many things.
• Missing validation or confidence ceilings for claims about reliability.

Practical Guidelines

• Prefer small, single-purpose tools with clear naming.
• Include example invocations and sample payloads in references/resources.
• Version tools when contracts change; keep changelog notes.

Cross-Skill Coordination

• Upstream: prompt-architect for clarity; cognitive-lensing for alternative designs.
• Parallel: skill-builder/skill-forge for directory and documentation alignment.
• Downstream: agent-creator/agent-selector using tool metadata; recursive-improvement for tuning reliability.

MCP Requirements

• Document required MCP servers and permissions; tag WHO=meta-tools-{session}, WHY=skill-execution for memory usage.

Input/Output Contracts

inputs:
  problem: string  # required problem statement
  environment: string  # optional runtime/environment notes
  constraints: list[string]  # optional constraints
outputs:
  tool_specs: list[file]  # definitions with contracts and examples
  validation_report: file  # tests executed and results
  composition_notes: summary  # how tools chain together

Recursive Improvement

• Use recursive-improvement when tools fail tests or integration friction is high; capture deltas between iterations.

Examples

• Build a schema-validator tool to sanitize API payloads before database writes.
• Compose a doc-generation toolchain combining linting, formatting, and preview steps.

Troubleshooting

• Flaky integrations → add retries, backoff, and circuit breakers.
• Schema drift → pin versions and add validation layers.
• Tool overlap → consolidate or specialize with clear naming and scope.

Completion Verification

• Tool contracts defined with inputs/outputs and failure modes.
• Validation evidence recorded; ceilings stated for reliability claims.
• Composition guidance documented for orchestration use.
• Dependencies and permissions disclosed.

Confidence: 0.70 (ceiling: inference 0.70) - Meta-tools SOP aligned to Skill Forge cadence with prompt-architect guardrails.