Skip to main content
Prompt Optimizer
Product
Docs
Pricing
Blog
Compare
Open main menu
Menu
Writing
Notes on prompt engineering, evaluation and routing
All
30
Content
6
Context Engineering
1
Prompt Engineering
4
AI Infrastructure
1
2026-09-19
Debug With Evidence: A Claude Skill for Root Causes
A free Claude skill that stops trial-and-error debugging: raw logs first, competing hypotheses, read-only probes, and a root-cause contract before any edit.
2026-09-19
How to Defend Against Prompt Injection: A Claude Skill
A free Claude skill that maps the input boundary, classifies injection attempts, scores severity, and applies a response plus four layers of defense.
2026-09-18
Why Agents Fail: A Claude Skill for Agent Prompt Design
Agents fail on tool mis-selection, context rot, and missing stop conditions. A free Claude skill applies a seven-stage protocol to design and review agent prompts.
2026-09-18
Stop Agent Context Bloat: The Context Cartographer Skill
A free XML-structured Claude Code skill that makes the agent write a task contract and gather minimum sufficient context before it edits any file.
2026-09-18
Right-Sizing Prompt Edits: A Claude Skill for Triage
Asked to clean up a prompt, models either rewrite everything or nothing. A free Claude skill scores the prompt first and allows only the edits its tier permits.
2026-09-18
How to Evaluate a Prompt: A Claude Skill for Testing
A free Claude skill that turns a prompt into a behavior contract and tests it with a matrix, deterministic checks first, and preserved evidence.
2026-09-18
Claude Prompt Optimizer Skill: Optimize Without Breaking It
A free Claude skill that classifies a prompt's context, judges its sophistication, and rewrites only what helps, keeping code, flags, and model IDs verbatim.
2026-09-18
Self-Improving Prompt Loops Fail When Nothing Reads the Output
My repair loop generated, scored, and committed better prompts for six weeks, but nothing read the result. Every state transition needs a reader.
2026-09-18
Should an Agent Spawn a Subagent? A Four-Question Test
A free Claude skill runs a four-question delegation test before an agent forks or spawns a subagent, then picks the shape, scopes the prompt, and sizes the wave.
2026-09-18
In Prompt Engineering, the System Around the Prompt Is the Prompt
Prompts are artifacts with histories, versions, and repairs. Five audits show whether the system around them works for a real user, not just in CI.
Older posts
showing 10 of 30