output-optimizer

name	output-optimizer
description	Optimize output token consumption through compact interpreter modes with controlled expansion when complexity, ambiguity, or risk requires more detail. Trigger: minimizing response verbosity while preserving clarity and correctness.
skillMetadata	{"author":"skilly-hand","last-edit":"2026-04-07","license":"Apache-2.0","version":"1.0.0","changelog":"Added a new portable output compression skill with deterministic interpreter modes and guarded detail expansion; reduces response token costs while preserving safety and clarity; affects response shaping workflows and catalog routing","auto-invoke":"When minimizing output verbosity or selecting compact communication modes","allowed-tools":["Read","Edit","Write","Glob","Grep","Bash","Task"]}

Output Optimizer Guide

When to Use

Use this skill when:

• You want compact responses to reduce output token usage.
• You need deterministic output formats for repeated workflows.
• You need concise communication without losing core clarity.
• You want controlled detail expansion only when risk or ambiguity requires it.

Do not use this skill for:

• Cases where the user explicitly asks for long-form teaching or narrative detail.
• Tasks that require extensive legal, medical, or compliance explanation by default.
• Situations where a fixed external output schema already overrides style choices.

---

Critical Patterns

Pattern 1: Activation and Precedence

Apply modes in this order:

1. If user writes mode: <name>, use that mode.
2. If no explicit mode, infer from phrasing:

• "keywords only" -> machine
• "yes or no" / "binary" -> binary-decision
• "json" / "structured output" -> json-compact
• "step by step, concise" -> step-brief
• "command style" / "minimal commands" -> neandertal
• "toon format" -> toon

3. If no strong signal, default to step-brief for human-readable compact output.

Explicit mode always wins over inferred mode.

Pattern 2: Mode Contracts

Mode	Contract	Token Profile
neandertal	Imperative command-like short phrases, no filler, minimal connectors.	Lowest human-readable
machine	Keywords only, grouped by labels, no prose sentences.	Ultra-low
step-brief	Numbered steps, each step max 3-4 short phrases.	Low with clarity
toon	Exactly 4 blocks: Title, Objective, Output, Next.	Low and stable
json-compact	Minimal stable JSON keys and short scalar values.	Low + parseable
binary-decision	yes or no plus one short reason.	Ultra-low for triage

Pattern 3: Complexity + Confidence Guard

Default to compact output. Expand only when:

1. Task complexity is moderate/high and concise output may cause mistakes.
2. Requirements are ambiguous and short output cannot preserve correctness.
3. Risk is elevated (security, production impact, irreversible operations).
4. User explicitly asks for more detail.

When expanding, keep structure compact and scoped to the needed clarification.

Pattern 4: Compression Rules

Always prefer:

• Specific nouns over long explanations.
• One-pass direct answer over repeated restatement.
• Bounded lists over paragraphs.
• Deterministic templates where possible.

Avoid:

• Polite filler and redundant transitions.
• Repeating the prompt unless needed for disambiguation.
• Verbose caveats when risk is low.

---

Decision Tree

User provided `mode: <name>`?                 -> Use explicit mode
No explicit mode, strong phrasing signal?     -> Infer mode from signal
No explicit mode and no signal?               -> step-brief
Task complexity/ambiguity/risk is high?       -> Expand within selected mode
User asks for detail/clarification?           -> Expand within selected mode
Otherwise                                     -> Keep compact output

---

Output Examples

Example 1: neandertal

Check logs. Find error. Patch file. Run tests. Report result.

Example 2: machine

status:blocked
cause:missing-env
action:set-token,retry

Example 3: step-brief

1. Open config file. Find auth block. Confirm token key.
2. Add missing key. Save file. Re-run command.
3. Verify success output. Capture result. Share summary.

Example 4: toon

Title: Auth Fix
Objective: Restore CLI login flow
Output: Config key added, login passes
Next: Run smoke check

Example 5: json-compact

{"status":"ok","mode":"json-compact","next":"deploy"}

Example 6: binary-decision

yes: tests pass on required suite

---

Prompt Patterns

These are prompt fragments, not terminal commands.

mode: neandertal
mode: machine
mode: step-brief
mode: toon
mode: json-compact
mode: binary-decision

explain in detail

---

Resources

• Mode protocol reference: references/mode-protocols.md
• Related complexity control: ../token-optimizer/SKILL.md