---

name: setup-memory-core description: "Bootstrap PyRIT setup, memory, registries, core models, and output helpers safely." disable-model-invocation: true metadata: disco-role: operating license: MIT

PyRIT setup, memory, registry, and output core

Use this sub-skill when a task needs to start a PyRIT process, choose a memory backend, load configuration safely, inspect or register framework components, retrieve stored conversations/results, or render PyRIT objects. It is for PyRIT 1.1.0.dev0 and assumes normal Python imports from the installed pyrit package.

Route here for

• Choosing between direct initialize_pyrit_async(...) and file-driven initialize_from_config_async(...).
• Creating no-secret quick sessions with InMemory or temporary/local SQLite memory.
• Understanding ~/.pyrit/.pyrit_conf, .env, .env.local, custom env_files, and Azure Key Vault references without committing secrets.
• Using CentralMemory after initialization and querying memory labels, conversations, scores, attack results, and scenario results.
• Inspecting PyRIT registries such as ConverterRegistry, TargetRegistry, ScorerRegistry, InitializerRegistry, and instance registries.
• Creating or checking core model objects (MessagePiece, Message, Score, AttackResult, identifiers, filters) needed by setup and memory workflows.
• Rendering attack results, conversations, scores, scorers, and scenario results with pyrit.output helpers and safe sinks.

Route elsewhere

• Prompt target credentials, target-specific constructor parameters, scorer selection, rate limits, and live service verification: targets-scorers.
• Converter classes, converter stacks, seed datasets, dataset YAML/data schemas, and modality conversion: converters-datasets.
• Running attacks, scenarios, executors, techniques, concurrency, retry policies, and interpreting attack outcomes: attacks-scenarios.
• pyrit_scan, pyrit_shell, pyrit_backend, backend REST server configuration, and scanner command flags: cli-backend-scanner.

Start sequence

1. Decide the initialization path and memory backend using references/setup-memory-registry.md.
2. Check exact signatures and model fields in references/api-reference.md before writing code.
3. If setup, registry, memory, or output fails, use references/troubleshooting.md before changing credentials, database files, or registry names.
4. For a no-secret local import/API smoke check, run scripts/setup_memory_smoke.py. The helper inspects installed APIs and can optionally construct lightweight in-memory/temporary objects; it never contacts external services.

Safety defaults

• Prefer InMemory for examples, tests, and one-off agent work where persistence is not needed.
• Prefer a caller-owned SQLite path under a temporary or project-approved directory when persistence is needed. Do not silently reuse an old database when labels/results look stale.
• Treat Azure SQL memory as optional and credentialed; use only when the task explicitly requires shared database state and the caller has supplied the required environment.
• Do not print, log, commit, or copy real .env values. Use .env.local for personal overrides and placeholders in examples.
• Avoid loading default ~/.pyrit files during no-secret checks; import and introspection are usually enough to validate package availability.