fpf-skillplanning-initialize-baseline

venikman's avatarfrom venikman

Creates an initial SlotFillingsPlanItem (A.15.3) baseline.

0stars🔀0forks📁View on GitHub🕐Updated Jan 10, 2026

When & Why to Use This Skill

This Claude skill automates the creation of a standardized 'Planned Baseline' for AI agent evaluation by generating initial SlotFillingsPlanItem objects. Following the FPF Pattern A.15.3, it ensures rigorous experiment tracking and reproducibility by enforcing explicit time selectors and defining essential slots for tool versions, task suites, and metric sets.

Use Cases

  • Benchmarking Initialization: Automatically generate a structured starting point for AI agent experiments to ensure all future performance is measured against a consistent baseline.
  • Reproducible Experiment Design: Create immutable snapshots of agent configurations, including specific model versions and task sets, to facilitate scientific comparison across different development cycles.
  • Automated Workflow Planning: Streamline the setup of complex agentic evaluation pipelines by programmatically defining the required infrastructure, metrics, and time-bound constraints.
namefpf-skill:planning-initialize-baseline
descriptionCreates an initial SlotFillingsPlanItem (A.15.3) baseline.
version0.1.0
allowed_tools[]

planning/initialize-baseline Kernel

Context

This skill implements FPF Pattern A.15.3 (SlotFillingsPlanItem). The goal is to create a "Planned Baseline" that future work is measured against.

Instructions

Generate a JSON object representing a SlotFillingsPlanItem with the following invariants:

1. Explicit Time (No "Latest")

The Γ_time_selector MUST be a specific point in time (e.g., point(2024-01-10T12:00:00Z)). You strictly forbidden from using "latest", "current", or "head".

2. Structure

Use the following schema:

{
  kind: "SlotFillingsPlanItem",
  bounded_context_ref: string, // e.g. "BC:AgentFS-Experiment-C1"
  path_slice_id: string,       // e.g. "P2W:bench-v1"
  Γ_time_selector: string,     // "point(...)"
  planned_fillings: [
    {
      slot_kind: string,       // e.g. "ToolVersionSlot" or "TaskSuiteSlot"
      planned_filler: string   // ByValue(...) or ByRef(...)
    }
  ]
}

3. Required Fillings

Include slots for at least:

  • ToolVersionSlot (Agent/Model config)
  • TaskSuiteSlot (Ref to tasks)
  • MetricSetSlot (Ref to metrics)