use-openhcs

Use OpenHCS

1. Call openhcs_health_check before relying on other tools. Report a bootstrap or stale-server failure instead of working around it.
2. Call openhcs_get_authoring_context with kind="first_use" when choosing or resuming an OpenHCS workflow, then request exactly the task context it names. For multisite assembly, registration, normalisation, segmentation review, or image-result QC, kind="image_analysis_workflow" is the canonical operating guide; follow it rather than reconstructing those rules from this skill.
3. Treat Python pipeline source as one complete PipelineDocument: pipeline_config plus ordered pipeline_steps. Never send a steps-only fragment, mirror configuration through a second argument, or strip source bindings from the reviewed document. UI-visible and headless routes use these same declarations even though their process and state owners differ.
4. Call openhcs_search_capabilities with task-relevant workflow, target, or text filters. Its current surface_profile, registry-owned workflow metadata, side effects, and security metadata—not a remembered tool list—decide whether to use a UI-visible or exposed headless route. Use openhcs_list_capabilities only when the complete selected surface is required. Before the first cold local function-catalogue search, discover catalog preparation capabilities and follow the existing responsive preparation procedure (openhcs_custom_function_workflow) on the intended existing endpoint. Observe its exact handle until READY; keep cold warming separate from ordinary 10-second tool observations. A timeout is not permission to restart or replay.
5. Use validated OpenHCS/CellProfiler benchmark examples as working contracts, especially the first-class Official30 corpus. Search the biological task plus OpenHCS Python, retrieve the exact openhcs_official30_benchmark_recipes section with max_chars=50000, and inspect its native reference, pipeline, input/channel assumptions, settings, and case-specific parity evidence. Benchmark parity is meaningful validation for that reference scope; it is not biological raw/overlay acceptance for a different assay. Record the example source and evidence tier before adapting it. CellProfiler image, object, measurement, relationship, and export semantics lower into the same OpenHCS declarations and runtime. For a compact, structured example of those fields, retrieve openhcs_official30_examplehuman_nuclei_recipe_card when the task involves ExampleHuman-style nuclei segmentation; its selected-value parity and unassessed new-assay QA are deliberately separate. When the task is unfamiliar or raw morphology conflicts with an example, follow pre-first-authorship learning retrieval (openhcs_analysis_learning) before choosing its method. Retrieve an applicable conditional lesson through existing knowledge search; benchmark defaults alone do not resolve contradictory raw evidence. Before authoring, retain the example-selection record described in the autonomous analysis strategy (openhcs_autonomous_analysis_strategy); reading this instruction is not evidence of retrieving a pipeline. Preserve failed checkpoints and continue authorised development; apply that guide's distinction between self-directed repair and external assistance, judging autonomous quality on the final frozen result rather than the first candidate. For an unfamiliar assay or an unresolved biological failure, use that guide's task-specific routes to image interpretation, preprocessing, segmentation diagnostics, measurements and recipe/error learning. Retrieve only the relevant guide; infer a provisional strategy from acquisition and matched raw channels before asking the user for missing information. These guides adapt Agentic-J's domain knowledge, not its Fiji command strings or model defaults.
6. Inspect real plate data and registered function declarations before authoring. If the microscope is unsupported, set PipelineConfig.source_bindings_config with LazySourceBindingsConfig from openhcs.core.config to filter files, extract metadata, name semantic sources, and project a virtual workspace; make each consuming FunctionStep select those aliases through its step-local source bindings. Never parse filenames inside processing functions. Reflect global or pipeline configuration with openhcs_describe_config_schema before setting non-obvious fields. Keep filesystem operations inside configured read and write roots. When importing an external metadata table, retrieve openhcs_image_sources#joining-an-external-metadata-table before declaring its join keys; plane-specific metadata and logical source-set metadata have different scopes. Before multi-channel photometry or consuming labels produced on another channel, retrieve openhcs_image_sources#plan-image-stacks-and-source-bound-labels-before-authoring; image-stack assembly and typed label input context are separate declarations.
7. Validate and compile before execution. Do not infer that source code, UI state, or an earlier validation result implies a current compiled plan.
8. Start read-only. Continue routine bounded work within existing task authorisation without asking again for each edit, execution or viewer action. Follow Task authorization in openhcs_architecture_quick_start for actions outside that authority; capability-registry metadata, path policy, declared confirmations and fresh revision/request tokens still apply.
9. Preserve the active ownership route. Discover the GUI bridge before UI tools and apply code/state changes with current tokens. Use headless tools only when their workflow group is exposed. Treat structured errors and recovery hints as authoritative; never bypass path policy, stale-process checks, compile requirements, or bridge authentication.