Bump inspect-ai from 0.3.220 to 0.3.223

[dependabot]

Bumps inspect-ai from 0.3.220 to 0.3.223.

Changelog

Sourced from inspect-ai's changelog.

0.3.223 (18 May 2026)

• Config: Add inspect log export-config command to export a run config from an existing log file.
• Anthropic: Skip thinking blocks when placing lookback cache_control.
• AsyncFilesystem: Add get_file() and exists() methods.
• Inspect View: Fix regression where switching task tabs would reload log, causing latency.

0.3.222 (16 May 2026)

• Scanners: Declare Scanner import in a way that's compatible with pyright type checking.

0.3.221 (16 May 2026)

• OpenAI: Add GPT 5.5 as computer use model and exclude 'chat' and 'instant' models from computer use.
• OpenAI Compatible: Parse OpenRouter-style reasoning_details in OpenAI-compatible responses.
• Anthropic: Capture extra_body fields from Message response.
• OpenRouter: Enable Anthropic prompt caching by default for openrouter/anthropic/* models.
• VLLM: Preserve dotted vLLM server arg keys.
• Bedrock: Drop unsupported sampling params for Claude 4.7+.
• Bedrock: Route top_k correctly for Nova models.
• SageMaker: Add prompt_logprobs support in chat mode via GenerateConfig, parse prompt logprobs from completion mode responses, enabling perplexity() and target_perplexity() scorers end-to-end.
• Model API: --adaptive-connections is now enabled by default (defaults to 100 per model connection).
• Model API: Cache lookup of openai and anthropic packages at sample initialization.
• Model API: Remove semaphore around calls to count_tokens() (they are already retried and gated by max_samples).
• Model Info: Cache model info database lookup results so that failed lookups don't repeat fuzzy model name search.
• Limits: Added suspend_token_limit() context manager for suspending token tracking and limit enforcement within a scope.
• Datasets: hf_dataset retries transient Hugging Face errors (rate limits, timeouts, Hub-unreachable cache misses) up to 3 times (5 in CI) with exponential backoff. Pass retry=False to disable.
• Datasets: Reject sample ids that collide under str() coercion.
• Datasets: Treat NaN from HuggingFace dataset as None is treated (converted to "").
• Datasets: Use HuggingFace revision in cache key for downloaded datasets.
• Datasets: Propagate hf_dataset(..., shuffle=True) to EvalDataset.shuffled.
• Tool Calling: Raise a ToolError if there is a null byte in command input.
• Scoring: Store and aggregate results for cancelled eval runs.
• Scoring: match(numeric=True) no longer matches digit-substrings (e.g. target 5 against 25); now correctly handles negative, decimal, and scientific-notation targets, and recognises unicode-formatted numbers (unicode minus, vulgar fractions like ½, Chinese numerals, fullwidth digits) in both targets and model output.
• Scoring: match(numeric=True, location="exact") is now strict — values like "5 some text" no longer match target "5".
• Analysis: Use score reducer in evals_df() column name when there are multiple reducers.
• Hooks: Cache list of registered hooks (invalidate cache on registry_add()).
• Config: Add --run-config option to inspect eval for single-file run configuration.
• Eval Set: Run Inspect Scout scanners over each task's logs as part of eval_set (CLI --scanner / ScannerConfig). Scans incrementally as logs land, reuses prior results across resumes, and renders progress alongside the existing eval view.
• Eval Set: Fail fast with "No inspect tasks were found at the specified paths." when a task spec resolves to nothing (e.g. uninstalled package); previously crashed with IndexError inside resolve_tasks after passing an empty task list to eval.
• Eval Set: Add score_display argument to eval_set() function.
• Eval Log: Preflight ETag check on S3 conditional write (required for S3 backends that don't implement conditional writes).
• Eval Log: Make log_file_info() robust to non-standard filenames; added log_file_info_async() / log_files_from_ls_async() so view-server header reads don't block the event loop.
• Imports: Delay importing heavier dependencies (e.g. s3fs, boto3, numpy, rich.markdown) for faster imports of inspect_ai module.
• Logging: INSPECT_PY_LOGGER_FORMAT env var (rich/plain/json) for non-TTY-friendly single-line console logs.
• Docker Compose: accept depends_on / pull_policy / privileged / shm_size / ulimits in ComposeService.
• Task Display: Honor terminal COLUMNS and LINES for dumb terminals.
• Validation: Reject unknown GenerateConfig fields with an error.
• Memory: Log condensing no longer retains unchanged JSON copies in long evals.
• Memory: Don't retain message lists in buffer DB (memory leak on long agentic samples).

... (truncated)

Commits

• 90a7b1c Update CHANGELOG for version 0.3.223
• df27f82 Bump to latest (#3970)
• 60c6326 AsyncFilesystem: add get_file and exists (#3964)
• 644f46f Merge branch 'kaifronsdal-fix/cache-control-skip-thinking'
• 61ad3ec changelog / lint
• c7a73dc Merge branch 'main' into fix/cache-control-skip-thinking
• c4a8f2f Add inspect log export-config command (#3959)
• a634db7 Skip thinking blocks when placing lookback cache_control
• eea5a68 Mount transcript search API from Scout when Scout is installed (#3947)
• 7b0f474 docs: document running vLLM solver and judge on separate servers (#3957)
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)