Roadmap - Foundry

The roadmap is an editorial view over the Foundry’s GitHub issues. The checked-in prose explains why each work area matters; live issue labels, native sub-issue relationships, titles, and states define what belongs here and are verified in CI. Near-term priority/mvp work appears first, followed by priority/v2 work.

At a glance

Work areas

#500 — Implement actual interviews.

Replace the interview path’s current placeholder behavior with a useful guided exchange that can discover a user’s workflow intent, constraints, evidence, and unresolved choices before producing the free-form summary consumed by downstream Molds.

Substeps

#377 — interview-to-freeform-summary is reviewed + pipelined but declares zero references

#67 — Research Nextflow test profile selection

Define how the Foundry chooses among Nextflow test profiles and cases so that translation starts from the smallest representative runnable path, records when several choices are meaningful, and asks the user only when evidence cannot settle the selection.

#66 — Research Nextflow test fixtures to Galaxy job inputs

Map Nextflow profiles, parameters, sample sheets, local or remote fixtures, and expected outputs into concrete Galaxy workflow-test job inputs, including collection construction and trustworthy datatype or hash assertions.

#59 — Research Nextflow process to Galaxy tool wrapper mapping

Ground tool-wrapper authoring in an explicit mapping from Nextflow process evidence to Galaxy inputs, outputs, command construction, containers, and optional behavior, including difficult cases such as shell fragments and processes that invoke several tools.

#267 — Source-agnostic Galaxy input collection-shape selection note (+ IWC survey to ground it)

Provide shared, corpus-grounded guidance for selecting Galaxy input collection shapes such as lists, paired collections, and sample sheets so each source-specific translation path can make the same decision from evidence rather than embedding its own heuristics.

#501 — Improved Support for Tool Development

Make custom Galaxy tool development a reliable branch of workflow construction by packaging authoritative upstream guidance, exposing the actual user-tool schema, validating generated wrappers mechanically, and keeping unsupported container or interface choices explicit.

Substeps

#318 — Revise how skills working on draft coordinate - more direct, more deterministic, more YAML

Move coordination state into the draft workflow itself so unresolved topology and tool-discovery outcomes become explicit YAML TODOs, the next unit of work can be selected deterministically, and the requirements ledger is reserved for obligations rather than general agent handoff.

#332 — Introduce a finalize-workflow step.

Add a distinct finalization phase after the draft becomes computationally complete, giving the Foundry one place to review the whole workflow, improve names and comments, resolve presentation concerns, and perform checks that do not belong in per-step implementation.

#365 — Foundry should use (or know about) job cache

Integrate Planemo and Galaxy job caching into repeated Foundry runs through a persistent project profile, expose cache-aware execution where the CLI permits it, and teach the harness to distinguish a legitimate reuse from rare false-green or crashed-tool failure modes.

#504 — Improve Pattern Pages

Turn Pattern pages into stronger executable references by showing data flow for multi-step operations and ensuring each pattern is backed by a small test workflow that demonstrates the claimed construction rather than relying on prose alone.

Substeps

#476 — Build a Pi-backed black-box Pipeline evaluation harness

Build a repeatable black-box harness that runs published Pipeline and skill artifacts through Pi workers, captures structured evidence, and separates exploratory refinement from conformance results that can support regression detection and credible release claims.

#491 — Add IWC-policy workflow review pipeline with conditional Foundry quality checks

Compose summarization, structural validation, execution, and one advisory review Mold into a Galaxy workflow review Pipeline grounded in accepted IWC policy, with additional Foundry quality checks applied only where the available evidence supports them.

#535 — Prove the IWC review pipeline on a real PR and accept its policy pin

Convert the structurally complete review Pipeline into scored evidence: land the upstream policy revision so the vendored review command is pinned to an accepted commit rather than a proposal, record how the Claude and Copilot review policies relate, and run all four phases against a real IWC pull request so the eval properties and harness gates are observed rather than asserted.

#492 — Add a checklist-driven Galaxy workflow maturation pipeline for IWC publication

Provide a lifecycle Pipeline that applies the accepted IWC contributor checklist to an existing Galaxy workflow, makes well-supported metadata and generalization improvements, preserves ambiguous decisions for users, and validates and reruns the result before presenting it as publication-ready.

Substeps

#496 — Implement the mature-galaxy-workflow-for-iwc Mold

#531 — Prove the IWC maturation pipeline with runnable fixtures and a .ga entry case

Close the evidence gap beneath the maturation Pipeline by authoring a runnable checklist fixture with real steps, outputs, and committed test data, walking it green and then deliberately red through all four phases with the failure attributed instead of any assertion weakened, and binding the lossless-normalization property to a committed .ga entry case.

#90 — Survey: Galaxy reporting and publishing patterns

Survey the IWC corpus to establish how Galaxy workflows select, label, group, and expose user-facing outputs, giving later reporting and publication work concrete patterns instead of inventing a generic output policy from first principles.

#306 — Pipeline construction for Loom/Orbit: a `loom` cast target that lowers pipelines into Loom plans

Add a Loom cast target that lowers Foundry Pipeline descriptions and their generated skills into executable Loom plans while preserving the architectural boundary in which the Foundry publishes portable knowledge and Loom owns stateful orchestration.

#499 — First Class CWL Support

Bring CWL to first-class status across the Foundry by exercising CWL-to-Galaxy end to end, tightening CWL summarization, producing the missing test-plan handoffs, and supporting CWL as a target alongside the paths already proven for Nextflow and free-form sources.

Substeps

#507 — Improved Structure and Schemas for Mold Artifacts

Strengthen the contracts passed between Molds by replacing loosely interpreted Markdown where stable structure has emerged, while retaining narrative handoffs where judgment and uncertainty matter and avoiding schemas that merely formalize unproven abstractions.

Substeps

#40 — Write IWC exemplar match schema

Define a traceable structured result for selecting the nearest IWC exemplar, including the chosen workflow, matching rationale, relevant structural features, evidence locations, and confidence for downstream template and implementation Molds.

#225 — Document Galaxy list-of-record collections for reference data

Research Galaxy’s record-shaped collection types as a possible representation for structured reference assets, clarifying when list:record is preferable to sample-sheet collections and grounding any recommendation in current Galaxy behavior and real workflows.

#308 — Design a Claude dynamic-workflow projection for Foundry Pipelines

Evaluate how Foundry Pipeline phase graphs, branches, loops, typed handoffs, and validation gates could compile into Claude dynamic workflow scripts without moving harness responsibility into the Foundry or duplicating orchestration semantics in another source format.