Token Budget Advisor

Offer users informed choice about response depth before answering. Control token usage, response length, and answer depth. Trigger phrases: token budget, tok...

Security Scan

This instruction-only skill is internally consistent with its stated purpose: it intercepts response flow to offer and remember user-chosen response-depth options and does not request extra credentials or install anything.

Purpose & Capability

The name and description (controlling token usage/response depth) match the SKILL.md instructions. It doesn't request unrelated binaries, environment variables, or config paths.

Instruction Scope

The instructions are focused on prompting the user for a depth option, estimating token usage heuristically, and then producing answers at that level. They require access to session context (to know if depth was already set) and give the agent some discretion about when to trigger, which is expected but somewhat subjective. No file I/O, external network endpoints, or secret exfiltration are requested.

Install Mechanism

There is no install spec and no code files — the skill is instruction-only, so nothing is written to disk and no external packages are pulled.

Credentials

The skill declares no required environment variables or credentials. The runtime instructions do not reference any secrets or config beyond session state, so requested access is proportionate to the stated function.

Persistence & Privilege

The skill asks the agent to 'maintain user's chosen level for rest of session.' This implies session-state persistence but does not request permanent or cross-skill configuration or always-on privileges. Users should verify how the host agent stores session preferences (in-memory vs persistent storage) if that matters to them.

Assessment

This skill looks coherent and low-risk: it will ask users to pick a response depth and remember that choice during the session, estimate token usage heuristically, and then answer accordingly. It does not request credentials or install code. Things to consider before enabling: test how often it triggers (its trigger rules are somewhat subjective), confirm how your agent persists session preferences (if persistent storage is a concern), and be aware token estimates are approximate since it uses a heuristic rather than a tokenizer.

Intercept response flow to let users choose depth before answering.

When to Use

• User mentions tokens, budget, depth, or response length
• User says "short version", "tldr", "brief", "exhaustive", etc.
• User wants to control detail level upfront

Do not trigger when: user already set depth this session, or answer is trivially one line.

Depth Options

Present before answering:

Input: ~[N] tokens | Type: [type] | Complexity: [level]

[1] Essential (25%)   -> ~[tokens]   Direct answer only
[2] Moderate (50%)    -> ~[tokens]   Answer + context + example
[3] Detailed (75%)    -> ~[tokens]   Multiple examples + alternatives
[4] Exhaustive (100%) -> ~[tokens]   Everything

Level Guide

Level	Length	Include	Omit
25%	2-4 sent	Direct answer, conclusion	Context, examples
50%	1-3 para	Answer + context + 1 example	Deep analysis, edge cases
75%	Structured	Examples, pros/cons, alternatives	Extreme cases
100%	No limit	Everything	Nothing

Shortcuts

If user signals depth, respond at that level immediately:

• "1" / "25%" / "short" / "brief" / "tldr" → 25%
• "2" / "50%" / "moderate" → 50%
• "3" / "75%" / "detailed" → 75%
• "4" / "100%" / "exhaustive" → 100%

Maintain user's chosen level for rest of session unless changed.

Estimation

Heuristic-based (no real tokenizer):

• Input tokens = estimate from prompt size
• Response window = input × complexity multiplier
• Accuracy ~85-90%, variance ±15%