Prompt Optimization

Prompt optimization

Todd of AIDLC optimizes prompts before they reach any AI provider. You get fewer wasted tokens and clearer, more reliable answers — on Copilot, Cursor, Claude, Devin, and Kiro alike.

On by default. Configure under Todd of AIDLC: Open Configuration → Workspace.

---

Why it matters

	Without optimization	With Todd of AIDLC optimization
Efficiency	Long chats and duplicate lines inflate every request	History trim, dedupe, file caps, merged system text
Quality	Models may ramble or repeat attached files	Response contract: goal first, minimal diffs, mode-specific rules
Multi-provider	Rules differ per vendor UI	Same Todd of AIDLC guidance on every provider pill

---

Algorithm (pipeline)

Before AgentRouter sends your message:

1. Normalize — trim whitespace; collapse excessive newlines (code fences preserved).
2. Dedupe — skip back-to-back identical user messages.
3. Prune — remove empty assistant placeholders.
4. Merge — combine duplicate Todd of AIDLC system guidance into one block.
5. Guidance — inject provider-agnostic instructions (efficiency + quality contract).
6. Mode hints — Ask / Agent / Spec+Agent each get targeted behaviour (plans, specs, no unsolicited edits).
7. Trim history — keep recent turns only (default 24 messages).
8. Context files — truncate large attachments (default 12 000 chars/file, head + tail kept).

---

Quality benefits

• Goal-first answers — less preamble and process narration
• Structured output — bullets for lists; short paragraphs otherwise
• Safer coding — Agent mode encourages small steps and self-check
• Spec+Agent — specs treated as authoritative acceptance criteria
• One clarifying question when blocked — reduces wrong guesses on security or data loss

---

Efficiency benefits

• Lower estimated tokens in on long sessions (see User Manual → Spending)
• Fewer duplicate charges from repeated user text or stacked system blocks
• Large context files remain useful without sending the entire file every turn

---

Settings

VS Code setting	Default
toddspect.promptOptimization.enabled	true
toddspect.promptOptimization.maxContextCharsPerFile	12000
toddspect.promptOptimization.maxHistoryMessages	24

---

Related

• Why Todd of AIDLC — product advantages
• Context and Specs — what gets attached before optimization
• Copilot Modes — Ask / Agent / Spec+Agent
• Repo doc: docs/prompt-optimization.md