Context Optimization

Audit and optimize the context that candid loads during reviews.

Why Optimize Context?

Every code review loads fixed context into the LLM: your Technical.md, config files, and decision register. This competes with tokens available for reading your actual code. Lean context means deeper code analysis.

Quick Start

/candid-optimize

This runs a full audit and presents recommendations interactively.

What It Checks

Token Budget Estimation

Shows a breakdown of how many tokens each context source consumes:

• Technical.md — your project standards
• Decision register — tracked questions and decisions
• Config files — project and user configuration

Flags when total fixed context is unusually high.

Technical.md Efficiency

Goes beyond /candid-validate-standards to analyze token efficiency:

• Verbose rules — rules that could be condensed without losing meaning
• Near-duplicate rules — rules with overlapping content across sections
• Low-signal rules — generic rules without project-specific references

Each finding includes estimated token savings and a condensed alternative.

Exclude Patterns

Scans your repo for files that waste review context:

• Generated files (*.generated.ts, *.g.dart)
• Build output (dist/, build/, .next/)
• Vendored code (vendor/, third_party/)
• Minified files (*.min.js, *.min.css)
• Lock files (package-lock.json, yarn.lock)
• Source maps (*.map)

Compares against your current exclude config and suggests missing patterns.

Decision Register

If the decision register is enabled:

• Recommends "lookup" mode when "load" mode register is large
• Suggests pruning old resolved entries beyond 100
• Detects duplicate questions

Config Tuning

Checks for common misconfigurations:

• Technical.md exists but no config.json
• Missing exclude patterns for detected generated/vendor files
• Large Technical.md without focus mode usage
• Merge target branches not configured

Usage

/candid-optimize                          # Full audit, interactive apply
/candid-optimize --dry-run                # Show recommendations only
/candid-optimize --apply-all              # Apply all optimizations
/candid-optimize --section technical-md   # Only analyze Technical.md
/candid-optimize --section excludes       # Only analyze exclude patterns
/candid-optimize --section register       # Only analyze decision register
/candid-optimize --section config         # Only analyze config tuning

Example Output

Context Budget Estimate
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Source                    Tokens     Budget
──────────────────────────────────────────────
Technical.md              ~2,400     ████████░░  62%
Decision register         ~1,200     ████░░░░░░  31%
Config files              ~100       ░░░░░░░░░░   3%
──────────────────────────────────────────────
Fixed context total       ~3,700

After applying optimizations, a before/after summary shows the token reduction.