An advanced Agent Skill designed for SREs, DevOps, and Systems Analysts. It reconstructs chaotic incident timelines, strictly separates observed facts from causal hypotheses, and subjects every assumption to 5 formal logical tests to prevent AI hallucination.
🎯 Why this Skill exists
Standard LLMs fail at incident analysis. They tend to "narrativize" the data, inventing causal links between events that merely happened close in time (correlation vs. causation). They also blindly trust the notes written by humans, even when those notes contradict the raw logs.
This Skill forces the AI to stop guessing. It acts as a cold, objective logic engine. If the data does not prove a root cause, the Skill will explicitly state: "Root cause not determinable from the provided data."
🧠 Core Concepts
To use this Skill effectively, you must understand how it categorizes information:
- Observed: The input explicitly states it. (e.g., "Log: HTTP 500 at 10:05").
- Inferred: Not stated, but follows necessarily from observed data. (e.g., "10:05 EST converted to UTC is 15:05 UTC").
- Hypothesis: A plausible explanation that the data does not prove. (e.g., "The database crashed due to memory exhaustion").
- Claim: A causal statement written inside the input by a human (e.g., "Note from Engineer: The deploy broke the server"). A Claim is NEVER treated as a fact. It enters the analysis as a hypothesis and must pass the logical tests.
⚙️ How it Works (The Workflow)
When triggered, the Skill runs a strict 6-phase process:
- Intake Check & Topology: Scans timezones, clock skews, gaps in logs, and explicitly stated system dependencies.
- Timeline Reconstruction: Normalizes timestamps to UTC and orders events (E1, E2, E3...) without inventing missing data.
- Facts vs. Hypotheses Matrix: Separates what actually happened from what people think happened.
- The 5 Logical Tests: Every hypothesis is tested for:
- Temporal Precedence (Did the cause happen before the effect?)
- System Proximity (Are the systems actually connected?)
- Alternative Explanations (What else could have caused this?)
- Negative Evidence (Is there data that proves this wrong?)
- Missing Links (What data do we need to confirm this?)
- Root Cause Assessment: Only survives if it passes the tests. If no hypothesis reaches "Medium" confidence, it says so.
- Risk & Action Plan: Predicts the next 2 most likely points of failure and gives a concrete 3-step mitigation plan.
🚀 Installation Guide
- Click the "Copy SKILL.md" button on the Agensi platform to download the file.