omp-flow-research

name	omp-flow-research
description	Investigate repository or external evidence within an omp-flow task Bundle, maintain source provenance as linked Concepts, and synthesize a justified direction.

OMP-Flow Research Gate

Preconditions

• The assignment identifies the Bundle root, bounded objective, entry Concept, output Concept, actor ID, and dispatch receipt.
• The entry contains a usable question or links to the current framing.
• Each independent question has an explicit output path, normally under research/.

Decide Research Scope

For each consequential question, state which provisional anchor assumption, principal contradiction, or practical decision its answer could change. Research that cannot affect the framing, a decision, or a verification obligation needs a concrete reason to consume time.

Choose and persist one of:

• Internal: repository code, tests, accepted context, history, existing patterns, and applicable durable knowledge discovered through the native omp-flow-wiki Skill.
• External: primary documentation and mature projects; repository clones are acquired into the ignored cache under .omp-flow/cache/repos/.
• Both: default for consequential architecture or unfamiliar domains.
• Skip: only when the user explicitly declines, the change is mechanical inside accepted knowledge, or existing evidence is sufficient. Record the concrete reason in a linked Concept.

Procedure

Before native dispatch, only Main/coordinator may dispatch, correlate operations/receipts, obtain or record human calibration, and choose a workflow transition. Those coordinator actions are inapplicable to an already-dispatched Researcher: it must not dispatch or self-redispatch, govern, calibrate, transition, or selectively reinterpret coordinator clauses. The Researcher still owns the complete bounded Research responsibility and assigned output: test the framing against practice, distinguish what evidence proves, does not prove, and merely makes possible, and report a materially changed principal contradiction as a Brainstorm-return signal rather than protect the earlier framing.

1. Split uncertainty into independently investigable questions and create descriptive output paths.
2. In Main/coordinator context only, dispatch Harness-native researcher agents when questions can be investigated independently. Pass Bundle root, role, objective, entry Concept, exact output path, actor ID, receipt, and the no-implementation boundary. Start one operation per native task item, then forward its complete returned assignment unchanged with id = actor_id = actorId and matching descriptor role. The strict v1 ompFlowDispatch descriptor remains the first non-blank line; do not decorate or reconstruct it.
3. Separate facts, interpretations, counter-evidence, unknowns, and recommendations. Actively seek the strongest evidence against the current framing, not only confirmation. Internal claims use file:line; external claims use stable source URLs and version/date where relevant.
4. Keep investigation in linked research Concepts. Do not copy passages merely to promote them between tiers.
5. Acquire useful repositories into the ignored clone cache and record the exact URL, revision, useful anchors, interpretation, and local relevance in one task-local Reference Concept.
6. Retain an exact attachment only when links plus revision are materially insufficient.
7. Write one or more synthesis Concepts comparing alternatives, risks, sources, and the justified direction. State whether the result confirms, revises, or falsifies the current anchor and what practical decision changes. Link the selected synthesis visibly from the Bundle index or relevant framing.
8. Recommend returning to Brainstorm when evidence materially changes the problem or principal contradiction; the main session updates human-owned framing. Otherwise explain why remaining uncertainty does not prevent the next decision and hand the selected synthesis path to design.

Exit Gate

• Important claims are evidenced or explicitly uncertain.
• Important sources have task-local Reference Concepts or a deliberate reason not to add one.
• One synthesis is visibly selected through authored prose and links.
• The synthesis exposes counter-evidence and its confirm/revise/falsify impact on framing and the next decision.
• The synthesis lets an Architect design without reconstructing research from chat.

In Main/coordinator context only, load omp-flow-design next. The already-dispatched Researcher returns the assigned synthesis/output and does not choose or perform that transition.

Red Flags

• No investigation means no design authority.
• Search snippets and model memory are not primary evidence.
• A cache clone is not task knowledge and is never accepted merely because it exists.
• Do not create paired content/metadata files or ask Python to reconstruct provenance.
• Do not hide disagreement to make the synthesis look complete.
• Do not continue investigation merely to complete an imagined evidence inventory after the material decision is supported or falsified.
• Do not require sortable filenames, fixed headings, or a closed Reference manifest.