agent-estimation_skill

Overview

This skill estimates AI agent work using the agent's native operational unit — the tool-call round — instead of human developer hours. It produces a module-by-module breakdown with round counts, risk coefficients, integration overhead, and a final wallclock conversion. The approach prevents human-time anchoring and yields reproducible, agent-centered estimates.

How this skill works

The skill decomposes the task into independent modules, assigns base rounds per module using calibrated anchors (boilerplate, moderate, exploratory, high-uncertainty), and applies a risk coefficient to each module. It sums effective rounds, adds integration rounds (10–20% of base total), and converts the total rounds to minutes using a configurable minutes_per_round parameter (default 3 minutes). The output is a clear table-style breakdown, a concise summary, and explicit biggest risks.

When to use it

• When asked to estimate or scope a coding task for an AI agent
• Before planning work that the agent will iterate on via tool calls
• When evaluating feasibility or comparing agent-driven options
• When you want reproducible estimates that avoid human-time bias
• When delivery depends on platform-specific setup or user-side actions

Best practices

• Always decompose into independent, testable modules before estimating
• Use the round anchors: 1–2 boilerplate, 3–5 moderate, 5–10 exploratory, 8–15 high-uncertainty
• Assign a risk coefficient (1.0–2.0) with a short rationale for each module
• Add integration rounds equal to 10–20% of base rounds to cover wiring and system interactions
• Only convert to wallclock at the end and set minutes_per_round based on user review burden

Example use cases

• Estimate time for adding a CRUD API endpoint plus tests and docs using rounds
• Scope integration of a third-party SDK with sparse docs and platform quirks
• Plan iterations for a data-pipeline module that requires exploratory schema mapping
• Compare two implementation approaches by their round counts and risk multipliers
• Audit previous agent estimates to identify human-time anchoring and rework them into round-based estimates