How saizy767/dynamic-prompt-core works
prompt-loop is a research pipeline for iteratively improving a binary classifier’s system prompt through error analysis and candidate scoring. Its main workflow runs a baseline, measures results, analyzes and selects thesis clusters, composes a prompt version, evaluates it against the active version, and accepts or rolls back. The graph preserves the documented optional teacher-refinement and scoring backends. Some subsystem links rely on README or SOURCE INDEX references; unsampled implementation details are not inferred.
Main entry points: run_cycle.py Cycle orchestrator · runner.py Baseline runner · comparator.py Version comparator · scoring Scoring adapters
Components 19
Cycle workflow
- Cycle orchestrator
src/dynamic_prompt_core/application/use_cases/run_cycle/run_cycle.py - Baseline runner
src/dynamic_prompt_core/application/use_cases/run_baseline/runner.py - Candidate selector
src/dynamic_prompt_core/application/use_cases/select_candidates/selector.py - Prompt composer
src/dynamic_prompt_core/application/use_cases/compose_prompt/composer.py - Version comparator
src/dynamic_prompt_core/application/use_cases/compare_versions/comparator.py
Analysis and refinement
- Classification metrics
src/dynamic_prompt_core/application/services/metrics/metrics.py - Thesis analyzer
src/dynamic_prompt_core/application/use_cases/analyze_theses/thesis_analyzer.py - Teacher refinement
src/dynamic_prompt_core/application/use_cases/refine_theses
Inference and scoring
- Candidate model
src/dynamic_prompt_core/domain/models/candidate.py - Scorer port
src/dynamic_prompt_core/application/ports/outbound/candidate_scorer.py - Classification policy
src/dynamic_prompt_core/application/services/classification_policy.py - Scorer factory
src/dynamic_prompt_core/infrastructure/llm/scoring/factory.py - Scoring adapters
src/dynamic_prompt_core/infrastructure/llm/scoring - Thesis extraction client
src/dynamic_prompt_core/infrastructure/llm/client.py
Prompt and run state
- Prompt version store
src/dynamic_prompt_core/infrastructure/storage/prompt_store.py - Run repository
src/dynamic_prompt_core/infrastructure/storage/run_repository.py
User interface
- CLI composition root
src/dynamic_prompt_core/interfaces/cli/main.py
Other components
- Researcher
- Inference server
Connections 19
- Researcher → CLI composition root: runs commands
README.md - CLI composition root → Cycle orchestrator: invokes cycle
src/dynamic_prompt_core/interfaces/cli/main.py - Cycle orchestrator → Baseline runner: runs versions
- Cycle orchestrator → Classification metrics: computes metrics
- Cycle orchestrator → Thesis analyzer: analyzes results
- Cycle orchestrator → Candidate selector: selects candidates
- Cycle orchestrator → Prompt composer: composes version
- Cycle orchestrator → Version comparator: compares versions
- Cycle orchestrator → Teacher refinement: may refine theses
- Baseline runner → Scorer port: scores candidates
src/dynamic_prompt_core/application/use_cases/run_baseline/runner.py - Baseline runner → Classification policy: classifies judgments
README.md - Baseline runner → Thesis extraction client: extracts theses
README.md - Scorer factory → Scoring adapters: builds backend
src/dynamic_prompt_core/infrastructure/llm/scoring/factory.py - Scoring adapters → Scorer port: implements scoring
src/dynamic_prompt_core/infrastructure/llm/scoring/factory.py - Baseline runner → Candidate model: uses candidates
src/dynamic_prompt_core/application/use_cases/run_baseline/runner.py - Version comparator → Baseline runner: runs candidate version
src/dynamic_prompt_core/application/use_cases/compare_versions/comparator.py - Cycle orchestrator → Prompt version store: activates accepted version
README.md - Cycle orchestrator → Run repository: persists run artifacts
- Thesis extraction client → Inference server: calls API