Workflow Template Catalog

June 19, 2026 · View on GitHub

Generated from the bundled Striatum workflow template catalog.

Workflow Shapes

Adjudicated constraint extraction (`adjudicated_constraint_extraction`)

An eight-phase productive-refusal loop: cross-examiners challenge a candidate synthesis, the adjudicator converts load-bearing objections into binding constraints, revision discharges each one, and final review typechecks discharge.

Recommended for: design and spec authoring; productive refusal; constraint-extracting multi-model panels
Default lane sets: multi_review, author_reviewer, local
Required options: workflow_id, artifact_root, topic Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Survey"]
  n1["Convener synthesis"]
  n2["Cross-examination"]
  n3["Adjudication (constraints)"]
  n4["Revision synthesis"]
  n5["Discharge review"]
  n6["Spec publication"]
  n7["Final review (typecheck)"]
  n0 --> n1
  n1 --> n2
  n2 --> n3
  n3 --> n4
  n4 --> n5
  n5 --> n6
  n6 --> n7
  n3 -.->|needs_revision| n1

Code change with bounded revision (`code_change`)

Draft, review, revise at most once if needed, then apply.

Recommended for: small code or docs edits that need an explicit review gate
Default lane sets: author_reviewer, single_agent
Required options: workflow_id, artifact_root Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Draft"]
  n1["Review"]
  n2["Apply"]
  n0 --> n1
  n1 --> n2
  n1 -.->|needs_revision| n0

Conversation (`conversation`)

N-turn, M-model alternating speaker conversation over the message bus.

Recommended for: model-to-model dialogue; agent-operator interviews; multi-turn reasoning loops
Default lane sets: author_reviewer, multi_review
Required options: workflow_id, artifact_root, topic Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

flowchart TD
  n0["Turn 1"]
  n1["Turn 2"]
  n2["Turn N"]
  n0 --> n1
  n1 --> n2

Cross-examination gate (`cross_examination`)

Require falsifying cross-examination and a rebuttal record before a finding or proposal can publish downstream.

Recommended for: finding readiness checks; proposal publication gates; challenge/rebuttal provenance
Default lane sets: multi_review, author_reviewer, local
Required options: workflow_id, artifact_root, topic Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Author draft"]
  n1["Cross-examiner"]
  n2["Adjudicator ledger"]
  n3["Commit"]
  n0 --> n1
  n1 --> n2
  n2 --> n3
  n2 -.->|needs_revision| n1

Custom safe block plan (`custom`)

Compose a workflow from known block kinds without raw workflow JSON.

Recommended for: advanced operators who need a graph not covered by a built-in shape
Default lane sets: custom
Required options: plan, workflow_id, artifact_root Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

No fixed graph preview.

Divergent ideation (`divergent_ideation`)

Widen a design space before narrowing it: diverge under cognitive frames (one per isolated branch), converge by scoring/clustering/trap-detecting, deepen the survivors, then synthesize. Frames are a curated, distortion-axis-tagged library (RFC 0129); branches round-robin across lanes for genuine multi-model divergence.

Recommended for: architecture and API design choices; naming; fuzzy-bug hypothesis classes; migration strategy; give-me-a-few-different-ways prompts
Default lane sets: multi_review, author_reviewer, local
Required options: workflow_id, artifact_root Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Frame problem"]
  n1["Diverge (frames)"]
  n2["Converge: score/cluster/trap"]
  n3["Deepen survivors"]
  n4["Final synthesis"]
  n0 --> n1
  n1 --> n2
  n2 --> n3
  n3 --> n4

Evidence-backed artifact (`evidence_backed`)

Produce claims with a support ledger and audit review.

Recommended for: artifacts whose claims need explicit evidence checking
Default lane sets: author_reviewer
Required options: workflow_id, artifact_root Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

flowchart TD
  n0["Draft"]
  n1["Support ledger"]
  n2["Evidence audit"]
  n3["Final review"]
  n0 --> n1
  n1 --> n2
  n2 --> n3

Falsification gate (`falsification_gate`)

Challenge a published proposal with falsifier artifacts, then gate downstream work on an adjudicator's collaboration ledger.

Recommended for: proposal readiness checks; assumption falsification; substance-gated static challenge/rebuttal
Default lane sets: multi_review, author_reviewer, local
Required options: workflow_id, artifact_root, topic Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Holder"]
  n1["Falsifiers"]
  n2["Adjudicator ledger"]
  n3["Commit"]
  n0 --> n1
  n1 --> n2
  n2 --> n3
  n2 -.->|needs_revision| n1

Fog-of-war review (`fog_of_war_review`)

Distribute disjoint spec fragments to reconstructor lanes that interrogate peers to recover the hidden constraints, then withhold the proposal until a full-spec judge scores reconstruction coverage through the substance gate (RFC 0094 §2 work-packet type sequencing).

Recommended for: partial-context reconstruction checks; constraint-coverage adjudication; deliberately-withheld-context design reviews
Default lane sets: multi_review, author_reviewer, local
Required options: workflow_id, artifact_root, topic Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

flowchart TD
  n0["Fragment distribution"]
  n1["Reconstruction (peer interrogation)"]
  n2["Coverage gate (judge ledger)"]
  n3["Proposal (withheld)"]
  n0 --> n1
  n1 --> n2
  n2 --> n3
  n2 -.->|needs_revision| n1

Human checkpoint (`human_checkpoint`)

Require owner judgment before downstream work proceeds.

Recommended for: runs that need an explicit operator decision gate
Default lane sets: author_reviewer, single_agent
Required options: workflow_id, artifact_root Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

flowchart TD
  n0["Analysis"]
  n1["Checkpoint"]
  n2["Apply"]
  n0 --> n1
  n1 --> n2

Implementation panel (`implementation_panel`)

Compare several implementation approaches with explicit scorecards, arbitration, dissent, and a final decision.

Recommended for: contested implementation choices; architecture trade-off resolution; high-risk design forks
Default lane sets: multi_review, author_reviewer
Required options: workflow_id, artifact_root Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Frame problem"]
  n1["Proposal A"]
  n2["Proposal B"]
  n3["Proposal C"]
  n4["Scorecards"]
  n5["Arbitration"]
  n6["Dissent review"]
  n7["Decision"]
  n0 --> n1
  n0 --> n2
  n0 --> n3
  n1 --> n4
  n2 --> n4
  n3 --> n4
  n4 --> n5
  n5 --> n6
  n6 --> n7

Iterated interrogating panel (`iterated_interrogating_panel`)

Two chained design+build loops, each fanning out to three independent lanes, synthesizing, then an interrogating panel review with a bounded needs_revision cycle.

Recommended for: high-stakes design-then-build work; patterns that need preserved-context interrogation; panel review with bounded re-work
Default lane sets: multi_review, author_reviewer
Required options: workflow_id, artifact_root Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

Example-only — not a workflow generate --shape value. Copy and adapt the example workflow at examples/iterated-interrogating-panel/workflow.json.

flowchart TD
  n0["Design (codex)"]
  n1["Design (claude_code)"]
  n2["Design (gemini)"]
  n3["Synthesis (interrogable)"]
  n4["Design review (threat_model)"]
  n5["Design review (ergonomics_dx)"]
  n6["Design review (devils_advocate)"]
  n7["Implement (interrogable)"]
  n8["Build review (threat_model)"]
  n9["Build review (ergonomics_dx)"]
  n10["Build review (devils_advocate)"]
  n0 --> n3
  n1 --> n3
  n2 --> n3
  n3 --> n4
  n3 --> n5
  n3 --> n6
  n4 --> n7
  n5 --> n7
  n6 --> n7
  n7 --> n8
  n7 --> n9
  n7 --> n10
  n4 -.->|needs_revision| n3
  n5 -.->|needs_revision| n3
  n6 -.->|needs_revision| n3
  n8 -.->|needs_revision| n7
  n9 -.->|needs_revision| n7
  n10 -.->|needs_revision| n7

Minimal bounded job (`minimal`)

One bounded job for a small report or starter artifact.

Recommended for: small reports; narrow inspections; first drafts
Default lane sets: local, single_agent
Required options: workflow_id, artifact_root Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Draft"]

Multi-phase workflow (`multi_phase`)

Run phase-scoped parallel tracks behind explicit synthesis gates.

Recommended for: large work split into ordered design, build, review, or release phases
Default lane sets: author_reviewer, multi_review, single_agent
Required options: workflow_id, artifact_root, phases Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

flowchart TD
  n0["Phase 1 track"]
  n1["Phase 1 synthesis"]
  n2["Phase 2 track"]
  n3["Phase 2 synthesis"]
  n0 --> n1
  n1 --> n2
  n2 --> n3

Multi-review synthesis (`multi_review_synthesis`)

Collect several independent reviews before a final recommendation.

Recommended for: productive disagreement; RFC or proposal review
Default lane sets: multi_review
Required options: workflow_id, artifact_root Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Review 1"]
  n1["Review 2"]
  n2["Synthesis"]
  n3["Final review"]
  n0 --> n2
  n1 --> n2
  n2 --> n3

Review and synthesis (`review`)

Draft, fresh review, then final synthesis.

Recommended for: proposal review; bug triage; documentation review
Default lane sets: author_reviewer, single_agent, local
Required options: workflow_id, artifact_root Support tier: supported — has a green RFC 0105 unattended-reliability fixture.

flowchart TD
  n0["Draft"]
  n1["Review"]
  n2["Synthesis"]
  n0 --> n1
  n1 --> n2

Synaptic prune (`synaptic_prune`)

After a forum closes, a post_dialog_hook (RFC 0094 §1) emits a prune packet before the participant lanes' preserved-context window releases; each still-live participant nominates one claim to retire, and any claim with >=2 votes is recorded in a durable collaboration_ledger injected as a negative preamble into future runs on the same topic.

Recommended for: retiring re-litigated claims; durable do-not-re-litigate records; post-forum consolidation
Default lane sets: multi_review, author_reviewer, local
Required options: workflow_id, artifact_root, topic Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

flowchart TD
  n0["Forum (post_dialog_hook)"]
  n1["Prune nomination"]
  n2["Tally (>=2-vote retire)"]
  n0 --> n1
  n1 --> n2
  n2 -.->|needs_revision| n1

Verification gate (`verification_gate`)

Make completion a CHECKED property, not a word a producer types. The builder ships a claim ledger; a real type:verify job runs striatum verifier run against sanctioned checks (builtin go-test/vet/build run with zero operator JSON, capped at ASSERTED) and mints receipts; the adjudicator gates the cleared release on the receipts. The sanctioned-check intent is in the verify lane's forbidden_paths (separation of duties), and VERIFIED is reserved for an external, operator-pinned-and-attested check (RFC 0141).

Recommended for: gating a release on machine-checked claims; completion provenance; stopping ideation/doc stages from claiming features the build never delivered
Default lane sets: author_reviewer, multi_review, local
Required options: workflow_id, artifact_root Support tier: experimental — no unattended-reliability gate yet (RFC 0105); expect to supervise.

flowchart TD
  n0["Build + claim ledger"]
  n1["Verify (run checks, mint receipts)"]
  n2["Adjudicate vs receipts"]
  n3["Commit cleared release"]
  n4["Finalize"]
  n0 --> n1
  n1 --> n2
  n2 --> n3
  n3 --> n4
  n2 -.->|needs_revision| n0

Lane Sets

Separate author and reviewer (`author_reviewer`)

Authoring jobs and review jobs bind to separate lanes.

Recommended for: independent review for a code or docs change
Required options: lanes.author.command, lanes.reviewer.command

Custom lane bindings (`custom`)

Advanced lane topology with every binding declared.

Recommended for: custom plans with explicit job-to-lane bindings
Required options: lanes, plan.job_lane_bindings

Local fixture lane (`local`)

Validation/scaffolding-only fixture lane set: its command sinks the packet (sh -c 'cat >/dev/null') and produces no artifact, so a run whose jobs declare expected_artifacts will park. Under RFC 0088 a claim requires an attached supervisor, so there is no human-types-the-artifact path -- use single_agent/author_reviewer/multi_review with real agent commands for runs that must produce artifacts.

Recommended for: tests; validation; starter scaffolds

Multiple reviewers (`multi_review`)

One author lane plus several reviewer lanes.

Recommended for: productive disagreement through multiple review postures
Required options: lanes.author.command

Single agent (`single_agent`)

One real agent session handles the whole workflow.

Recommended for: small low-risk work; early adoption
Required options: lanes.agent.command

Role Packs

Implementation panel roles (`implementation_panel_roles`)

Independent proposal authors, a scorekeeper, an arbitrator, and a dissent reviewer for explicit trade-off resolution.

Recommended for: implementation_panel; architecture_tournament
Default shapes: implementation_panel
Roles: problem_framer, proposer_a, proposer_b, proposer_c, scorekeeper, tradeoff_ledger, arbitrator, dissent_reviewer, principal_decider

Incident response roles (`incident_response`)

Reproduction, root-cause, fix-planning, verification, and retrospective viewpoints for incident closure.

Recommended for: incident follow-up; workflow failure analysis
Default shapes: review, evidence_backed
Roles: reproducer, root_cause_analyst, fix_planner, verifier, retrospective_author

Release readiness roles (`release_readiness`)

Release, documentation, migration, smoke-test, and rollback viewpoints for shipping decisions.

Recommended for: release readiness reviews; migration releases
Default shapes: review, multi_review_synthesis
Roles: release_manager, docs_reviewer, migration_reviewer, smoke_verifier, rollback_reviewer

Adversary Packs

Maintainer cost pressure (`maintainer_cost`)

Challenge approaches on long-term ownership, migration burden, and support cost.

Recommended for: architecture choices; large refactors
Default shapes: implementation_panel, code_change
Postures: maintainability, migration_risk, reversibility

Operator ergonomics pressure (`operator_ergonomics`)

Challenge approaches on setup clarity, repeated-use friction, recovery behavior, and local-first operation.

Recommended for: operator tools; workflow catalog entries
Default shapes: implementation_panel, review
Postures: operator_experience, recovery, documentation

Security and privacy pressure (`security_privacy`)

Challenge approaches on capability scope, data minimization, secret handling, and persistence boundaries.

Recommended for: MCP surfaces; artifact export; daemon methods
Default shapes: review, evidence_backed
Postures: security, privacy, capability_scope