A. Add a research-planning agent before the initial search phase to decompose every topic into detailed subquestions.

B. Have the coordinator look for general gap indicators in the analysis output and run additional searches without repeating the analysis stage.

C. Require the analysis agent to return specific evidence gaps to the coordinator, which launches targeted searches and invokes analysis again until the defined coverage criteria are satisfied.

D. Have the synthesis agent assign confidence scores to each report section and flag insufficiently supported sections for manual review.

A. Max tokens

B. Temperature

C. Context window

D. Prompt length

A. Increase randomness.

B. Remove formatting instructions.

C. Specify exact column names.

D. Omit examples.

A. Return fields with confidence scores, plus a request_review boolean computed using your tested confidence thresholds, along with a review_reasons array explaining which fields triggered review.

B. Compute an aggregate extraction_quality score across all fields and return it alongside the extracted values. Include a text summary describing the overall extraction reliability.

C. Return fields with their raw confidence scores and add detailed few-shot examples to your system prompt demonstrating how to interpret different confidence ranges and when to request human review.

D. Return fields organized into verified and needs_verification objects based on confidence thresholds.

A. Increase temperature to 1.0

B. Use Retrieval-Augmented Generation (RAG)

C. Disable system prompts

D. Shorten every prompt

A. Whether your result-processing system can handle reviews arriving in a different order from the order in which they were submitted.

B. Whether reducing per-review latency from 30?0 seconds to near-instantaneous delivery matters to your workflow.

C. Whether each review can be structured as a single request without multi-turn refinement.

D. Whether review feedback arriving up to 24 hours after pull-request creation remains actionable.

A. Add cache-prewarming requests with max_tokens: 0 at the beginning of every batch.

B. Move the cache_control breakpoint from the shared system prompt to each migration script so similar code patterns can be reused.

C. Configure the cache breakpoints to use the extended one-hour TTL instead of the default five- minute TTL.

D. Split the 200 requests into ten sequential batches of 20, submitting each batch only after the previous batch completes.

A. Keep --system-prompt and add --allowedTools "Read,Glob,Grep", because non-interactive -p mode otherwise disables filesystem tools.

B. Remove --system-prompt entirely and place the review instructions in a CLAUDE.md file, because - -system-prompt is incompatible with tool use under -p.

C. Replace --system-prompt with --append-system-prompt and explicitly instruct Claude to inspect related repository files whenever broader context is needed.

D. Stop piping the diff through standard input and embed it inside the prompt, so Claude Code treats the invocation as an agentic session.

A. Claude is malfunctioning.

B. The API returned cached responses.

C. Retrieval failed.

D. Large language models are probabilistic.

A. Systematically read all 200 files to create a complete function inventory before writing any tests, ensuring the testing plan accounts for every function before beginning.

B. Use Glob and Grep to map codebase structure, identify heavily-coupled modules, create a prioritized plan for high-impact areas, and revise as dependencies are discovered.

C. Create a fixed testing schedule upfront based on directory structure, allocating equal effort to each top-level directory regardless of code complexity or business importance.

D. Start writing tests for the first module alphabetically, using test failures and imports to discover related files organically.

A. Add server-side validation that checks whether order_id exists before attempting the refund and returns an error when it does not.

B. Change tool_choice from auto to any so Claude must call a tool on every turn.

C. Preprocess customer messages to extract any mentioned order identifiers and inject them into the conversation before sending the request to Claude.

D. Update the process_refund tool description to state explicitly that order_id must come from a successful lookup_order result and must never be assumed, inferred, or invented.

A. Implement retry logic with exponential backoff in your MCP server for all errors, returning to the agent only after retries are exhausted.

B. Add few-shot examples to the system prompt demonstrating how to interpret error message patterns and select appropriate responses for each.

C. Create an analyze_error MCP tool the agent calls after any failure to determine the error category and recommended action.

D. Enhance error responses with structured metadata: include errorCategory (transient/validation/permission), isRetryable boolean, and a description of what caused the failure.