Failure Postmortem

Overview

Systematically captures failure data. When a tool fails or a plan is abandoned, this skill logs the event to a persistent registry (POSTMORTEMS.md), forcing an synchronous analysis of why it happened (Root Cause) and how to prevent it (Preventive Rule).

Trigger Phrases

• log failure
• postmortem
• why did this fail
• record error

Inputs

| Parameter | Type | Required | Default | Description | |-----------|------|----------|---------|-------------| | --Command | string | Yes | - | The command or action that failed | | --Error | string | Yes | - | The error message or exception trace | | --Context | string | No | General | What was the agent trying to do? |

Outputs

1. Log Entry

Appends to .gemini/antigravity/.forensics/POSTMORTEMS.md:

## [2026-01-16 10:00:00] Failure Analysis
**Context:** Deployment
**Command:** `npm build`
**Error:** `Heap out of memory`

### 🕵️ Analysis
1. **Root Cause:** Node processes default to 512MB RAM.
2. **Preventive Rule:** Always set NODE_OPTIONS="--max-old-space-size=4096" for builds.

Preconditions

1. .forensics directory exists (auto-created).
2. PowerShell 5.1+ or Core 7+.

Safety/QA Checks

1. Persist: Ensures the log file is always writable.
2. No Data Loss: Appends only, never overwrites previous logs.

Stop Conditions

| Condition | Action | |-----------|--------| | Disk Full | Fail gracefully (stdout only) |

Implementation

See scripts/log_failure.ps1.

Integration with Other Skills

1. Agent Catch Block: On exception -> Call log_failure.ps1.
2. Agent Step: Read the output path.
3. Agent Step: "Thinking" block -> Fill in the "Analysis" section mentally or in next turn.