skills.rest
phrazzld
phrazzldCOMMUNITY
Explore all skills by this author →
GitHub

calibrate

Diagnose agent mistakes and guide harness improvements during runs.

13|3|Updated Mar 2, 2026
Search Tags:#diagnostics#calibration#quality-assurance#postmortem#harness#instruction-design
One-click install
npx skills add https://github.com/phrazzld/spellbook --skill calibrate-phrazzld
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: calibrate
Source: https://github.com/phrazzld/spellbook/tree/main/skills/calibrate
Command: npx skills add https://github.com/phrazzld/spellbook --skill calibrate-phrazzld

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

The calibrate skill guides teams to diagnose why an agent made a mistake and fix the harness before addressing the code, reducing recurrence and improving reliability.

Core Features & Use Cases

  • Pause and capture what happened, what was expected, and impact to anchor the investigation.
  • Load context from diagnostic skills and references to trace decisions and identify root causes.
  • Classify and escalate fixes to the highest reliable level (type system, lint, hooks) to prevent recurrence.

Quick Start

Describe the incident and run the /calibrate workflow to generate a Calibration Report.

Frequently Asked Questions about calibrate

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I diagnose agent mistakes mid-session and fix the harness?▼

To diagnose agent mistakes, pause the session to capture what happened, load diagnostic context to trace decisions, and escalate fixes to the type system, lint rules, or hooks to prevent recurrence.

What is the best way to prevent my agent from repeating the same error?▼

The best way to prevent repeated errors is to fix the harness by prioritizing type-system constraints, lint rules, and harness checks over patching the code directly, ensuring structural reliability.

How does root-cause analysis for agent failures work during an active run?▼

Root-cause analysis works by capturing the incident's expected behavior and impact mid-session, loading context from diagnostic skills and references, and classifying the harness gap that caused the failure.

Why should I fix the harness before addressing the code when an agent fails?▼

Fixing the harness before the code reduces recurrence by enforcing reliability at the highest level, using type systems and hooks to structurally prevent the agent from making the same mistake again.

Do I need specific diagnostic skills and references to trace agent decisions?▼

You need diagnostic skills and references to load context during a run, which allows you to trace the agent's decisions, identify root causes, and generate a structured calibration report.

What is a calibration report for harness issues and when do I need one?▼

A calibration report is a structured output generated by describing an incident, used when an agent makes a mistake mid-session to identify harness gaps and guide improvements across skills and hooks.

Related Skills

Explore More Software Engineering Skills→
redhat-etredhat-et
9

incident-investigator

Orchestrate logs-fetcher, root-cause-analysis, and context-fetcher to investigate incidents.

Official
Intermediate
hung-phanhung-phan

postmortems

Facilitate structured blameless incident postmortems with automated action item tracking.

Community
Advanced
paulinghampaulingham
1

harness-audit

Scan ~/.claude/ configuration to flag orphan hooks, missing skills, invalid JSON, and stale agents.

Community
Advanced
Pepsi1978Pepsi1978
1

resilient-bugfixing

Automate bugfixing with root-cause analysis, defensive patching, and validation.

Community
Advanced