llm-cost-optimizer

Use proactively whenever LLM API costs come up -- or should. Triggers include: 'my AI costs are too high', 'optimize token usage', 'which model should I use', 'LLM spend is out of control', 'implement prompt caching', 'we're about to launch an AI feature', 'build me an AI endpoint'. Don't wait for an explicit cost complaint -- if someone is building an AI feature, designing an LLM endpoint, or choosing between models, cost architecture belongs in the conversation. Apply immediately when any of these are true: a system prompt appears that exceeds a few hundred tokens, all requests are hitting the same model, max_tokens is not set, or no per-feature cost logging exists. NOT for RAG pipeline design (use rag-architect). NOT for improving prompt quality or effectiveness (use senior-prompt-engineer).

Quality

80%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

—

The risk profile of this skill

Quality

Content

75%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured, actionable decision framework that respects Claude's intelligence and includes validation/recovery loops. The main gaps are minor redundancy across tables and the absence of any copy-paste code/config artifacts for the prescribed techniques.

Suggestions

Add a minimal executable artifact or two (e.g., a per-request logging JSON schema and an Anthropic cache_control code snippet) to lift actionability from specific guidance to copy-paste ready.

De-duplicate the flagging guidance that appears in both Mode 2 and the Proactive Flags table by referencing one from the other.

Add an explicit rollback/validate-impact-before-scaling checkpoint to the Mode 2 ROI-ordered flow so each numbered step has a clear feedback gate.

Dimension	Reasoning	Score
Conciseness	Lean and dense with no basic-concept filler (e.g., 'AI API costs are engineering costs. Treat them like database query costs'), though some flagging content recurs across Mode 2 and the Proactive Flags table and the Anti-Patterns table partly restates Mode 2.	4 / 5
Actionability	Highly specific guidance ('Log per-request: model, input tokens, output tokens, latency...', 'cosine similarity >0.95 = safe to serve cached response', 'soft alerts at 80% of limit'), but it stops short of copy-paste code/config blocks (no cache_control snippet, logging schema example, or routing classifier code).	4 / 5
Workflow Clarity	Clear ROI-ordered sequencing ('Don't skip ahead -- measure impact at each step') with feedback loops in the Failure Modes table ('Restore compressed section', 'Fall back to rule-based routing'), though the main Mode 2 flow lacks an explicit rollback/validate-before-scaling checkpoint at each numbered step.	4 / 5
Progressive Disclosure	No bundle files exist and all content is inline in a single well-organized SKILL.md with clear headers and easy navigation; a couple of large inlined tables (Anti-Patterns, Output Artifacts) could plausibly be externalized, keeping it just below a 5.	4 / 5
	Total	16 / 20 Passed

Description

86%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description with comprehensive natural trigger phrases and excellent distinctiveness via explicit exclusions. The 'what' could be made slightly more concrete within the description itself rather than relying on the body.

Suggestions

Move a one-line concrete capability summary (e.g., 'Applies model routing, prompt caching, output-length control, and prompt compression to cut LLM API spend') from the body into the description so the 'what' is explicit alongside the triggers.

Dimension	Reasoning	Score
Specificity	Lists several specific signals and triggers ('system prompt... exceeds a few hundred tokens', 'max_tokens is not set', 'implement prompt caching'), though the concrete capability actions (routing, caching, compression) are more fully enumerated in the body than in the description itself.	4 / 5
Completeness	The 'when' is exceptionally explicit ('Use proactively whenever LLM API costs come up', 'Apply immediately when any of these are true', plus 'NOT for...' exclusions), but the explicit 'what' capability list leans on the body rather than the description.	4 / 5
Trigger Term Quality	Comprehensive coverage of natural user phrases with synonyms and varied phrasings ('my AI costs are too high', 'optimize token usage', 'LLM spend is out of control', 'build me an AI endpoint') plus technical triggers.	5 / 5
Distinctiveness Conflict Risk	Explicit boundary guidance ('NOT for RAG pipeline design (use rag-architect)', 'NOT for improving prompt quality or effectiveness (use senior-prompt-engineer)') carves a clear, distinct niche with minimal conflict risk.	5 / 5
	Total	18 / 20 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: alirezarezvani/claude-skills
Path: engineering/llm-cost-optimizer/skills/llm-cost-optimizer/SKILL.md
Commit: 19392f7

Reviewed: ~6 hours ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.