Anti-Patterns

The Anti-Patterns page is the core improvement engine of AI Engineer Coach. It runs a set of detection rules across your session data and produces scored findings with actionable recommendations.

Rule Engine

Anti-pattern detection is powered by an editable rule engine. Every detector is a self-contained markdown file with YAML frontmatter and a Detection Logic block written in a small domain-specific language (DSL). AI Engineer Coach ships 45 built-in rules spanning prompt-quality, session-hygiene, code-review, tool-mastery, and context-management. You can:

• Edit a rule's thresholds, description, and detection logic directly from the Rule Editor.
• Write a brand-new rule in natural language and let the AI builder scaffold the markdown and DSL for you.
• Live-test any rule against your own data, including threshold sliders for quick tuning.
• Drill into a rule coverage heatmap showing how each rule triggered across your workspaces.

Score Categories

Five practice categories are evaluated, each scored from 0 to 100:

Prompt Quality

Measures how well you provide context to the AI. Representative rules:

• Missing File Context -- Requests that do not reference files with #file or have open editors
• Lazy Prompting -- Short, unspecific requests that produce generic responses
• Caps Lock Rage -- Requests written mostly in CAPS indicating frustration
• Profanity / Hostile Language / Frustration Signals -- Sentiment-based rules that highlight breakdowns in the human-AI loop

Session Hygiene

Evaluates how you manage your coding sessions:

• Mega Sessions / Abandoned Sessions / Session Drift -- Flag sessions that run too long, stop abruptly, or drift across unrelated topics
• Late-Night Coding / Weekend Overwork -- Detect overwork patterns
• Repeated Prompts -- Surface near-duplicate prompts that suggest your context isn't landing

Code Review

Assesses how carefully you handle AI-generated output:

• Auto-Approve Terminal -- Terminal commands being auto-executed without a devcontainer
• Speed Accept / Copy-Paste Blindness -- Accept-heavy patterns that suggest minimal review
• YOLO Mode / No Devcontainer -- Unsafe execution sandboxes

Tool Mastery

Evaluates how broadly you use available features:

• No Slash Commands / No Plan Mode / Agentic No Tools -- Underused productivity features
• Premium Waste / Model Overreliance -- Using expensive models for simple questions
• No Skills / No Custom Instructions -- Gaps in your context-engineering setup

Context Management

Evaluates how efficiently your sessions use the context window:

• Context Window Saturation -- Sessions where the context window is nearly full
• Compaction Storms -- Frequent auto-compactions indicating sessions that run too long
• Runaway Growth -- Context size growing steadily without resolution

Findings

Each finding includes:

• PROBLEM -- A description of what was detected, with specific counts
• ACTION -- A concrete recommendation for how to fix it
• Examples -- Expandable section showing real examples from your sessions

Trends

Week-over-week (WoW) and month-over-month (MoM) trend indicators appear on each score card, so you can see whether your practices are improving or regressing.