HERO

2026-02-18-agent-failure-taxonomy

One-line value: Use 2026-02-18-agent-failure-taxonomy as an execution-ready playbook, not a note.

When to use: Use this page when you need to execute this workflow in one focused session.

QUICK RESULT

If you only do one thing → complete the first checklist pass and publish one usable draft/output today.

ACTION CHECKLIST

• [ ] Clarify the exact output and success metric before starting.
• [ ] Gather required inputs from one trusted source only.
• [ ] Execute the workflow in sequence without adding side tasks.
• [ ] Run one quality check and fix the highest-risk issue first.
• [ ] Save the final result with a short reuse note.

EXAMPLE / DEMO

Before: Notes are scattered and decisions are unclear.

After: Inputs are structured, steps are executed, and the output is ready to use immediately.

WHY IT WORKS

• Converts vague intent into an explicit sequence.
• Emphasizes shipping one validated result fast.
• Creates repeatability for future runs.

NEXT ACTION

• Run this checklist on one live task now; keep scope to a single measurable outcome.

Source notes (kept for context)

Taxonomy (starter)

• CONFIG: wrong paths, missing secrets, wrong env

• INPUT: malformed order, missing required fields

• TOOL: external dependency down, API changes

• LOGIC: bug in workflow, wrong branching

• DELIVER: output missing, wrong target dir

• QUALITY: output exists but fails acceptance checks

How to use it

1. Every FAIL gets exactly one primary class.

2. Record evidence (log line, file path, URL).

3. Fixes must include a guard so it doesn't recur.