Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+21 moreClaude CodeCursorCodex CLIMuseDots
Manus
Grok
Grok Bot+18 agentsAutoresearch Optimizer: Metric-Driven Iteration Agent
A controlled optimization framework that improves prompts and configs through iterative, measurable experiment loops.
$23.99
Autoresearch Optimizer: Metric-Driven Iteration Agent
Example session with this skill installed
I want to optimize a RAG system prompt. The baseline accuracy is 65%. I want to test if adding a 'think step-by-step' instruction improves accuracy without increasing latency by more than 10%.
- Read your context and instructions
- Compiled the autoresearch optimizer
I have initialized the optimization loop for your RAG prompt. Below is the experiment contract and the first iteration details.
Optimization Experiment: RAG Accuracy Improvement
Executive Summary
This experiment tests the addition of a Chain-of-Thought instruction to the baseline RAG prompt. We are measuring the impact on factual accuracy while monitoring latency as a primary guardrail.
Experiment Contract
- Objective: Increase factual accuracy in RAG responses.
- Primary Metric: Accuracy score (0.0 - 1.0) via a fixed 50-question test set.
- Guardrail: Latency must not increase by >10% over baseline.
- Acceptance Rule:
KEEPif Accuracy increases by >5% and Guardrail passes.
Baseline Status
- Version:
rag-prompt-v1.0 - Metric Score: 0.65
- Avg Latency: 1.2s
- Status:
MEASURED
Iteration EXP-0001: Chain-of-Thought Hypothesis
Hypothesis: Adding "Think step-by-step before answering" will improve reasoning quality and factual retrieval alignment.
- Candidate Diff:
- Answer the user question based on the context provided.
+ Think step-by-step about the context provided, then answer the user question.
Evaluation Results
| Case ID | Baseline Result | Candidate Result | Delta | Latency (s) |
|---|---|---|---|---|
| TC-001 | Fail | Pass | +1 | 1.4 |
| TC-002 | Pass | Pass | 0 | 1.3 |
| TC-003 | Fail | Fail | 0 | 1.3 |
| Aggregate | 0.65 | 0.72 | +0.07 | 1.32 |
Decision: KEEP
Rationale: Accuracy improved by 7% (exceeding the 5% threshold). Latency increased by 10% exactly (1.32s vs 1.2s), meeting the maximum allowable limit.
Next steps
- Promote
candidate-v2to the new baseline. - Formulate a new hypothesis focused on reducing the 10% latency overhead introduced in this round.
- Test a condensed version of the instruction to see if accuracy gains hold with lower compute cost.
Connects securely to your tools. The creator never sees your data.
About this skill
Iterative trial and error is a slow, expensive way to optimize prompts or configurations. Without a controlled process, you lose track of which changes actually drive performance and which ones just introduce regressions. This skill transforms haphazard tweaking into a rigorous optimization loop by enforcing scientific discipline on every iteration.
What it does
- Standardizes metrics by defining exact measurement methods, evaluation sets, and acceptance rules before any modification occurs.
- Isolates variables to ensure only one meaningful change is tested at a time, making results interpretable and reproducible.
- Establishes baselines by recording starting performance across per-case results, aggregates, and latency metrics.
- Enforces guardrails to prevent optimizing for a single metric while silently degrading quality, safety, or cost.
- Maintains audit trails through a structured experiment schema that logs every hypothesis, diff, decision rationale, and outcome.
How it works
- Define the contract by specifying the objective, primary metric, evaluation protocol, and hard budget limits.
- Freeze the baseline by measuring the current artifact under fixed conditions to set a performance floor.
- Execute iterations where a single hypothesis is generated, a candidate diff is applied, and results are compared against the baseline.
- Apply decision rules to keep validated winners, reject regressions, or flag inconclusive results for human review.
Frameworks & tools
This skill is platform-agnostic and works with any system requiring iterative refinement, including LLM prompts, model configurations, code snippets, and automated workflows.
Why this beats prompting it yourself
Manual prompting often leads to "whack-a-mole" fixes where one improvement breaks three other things. This skill prevents regression by requiring fair, deterministic evaluation and formal rejection of any change that fails predefined guardrails.
Use cases
- Prompt engineering to increase factual accuracy while staying within token budgets.
- Hyperparameter tuning for agentic workflows to reduce error rates and latency.
- Code optimization to improve execution speed without compromising security or correctness.
- System configuration to find the optimal balance between throughput and API costs.
Known limitations
Does not perform the actual execution of external code or API calls. Requires a stable evaluation dataset and measurable metrics to function effectively.
Requires an established evaluation dataset. Cannot execute external code or live API calls independently. Metrics must be measurable and provided by the user or an external environment.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 14 days ago