Skip to content

Workflow Library ​

Agentweaver ships seven reusable workflow definitions in packages/Agentweaver.Squad/Catalog/Resources/workflows/. A blueprint declares a workflows set; the coordinator selects for process fit, with explicit overrides. WorkflowRegistry also loads the built-in default and project-authored definitions.

Workflow index ​

IDPurposeImportant gates
software-deliveryPlan and implement softwareQA test, RAI, rubberduck, Build/Test, human review
bug-fixTriage and fix a defectQA verification, RAI, Build/Test, human review
content-authoringResearch, draft, edit and publish textRAI, human review
pm-discoveryResearch and synthesize a product proposalHuman review
agent-evaluationRun and report an agent evaluationRAI safety
incident-responseInvestigate, mitigate and document an incidentHuman review
infra-opsPlan and review infrastructure changesValidation, RAI, infrastructure review, human review

These are authored workflow graphs, not deployment topologies. Prompt steps and verdict-producing gates are distinct. The catalog no longer contains a standalone code-review identity; review remains a capability inside workflows.

The YAML definitions do not author Merge, PR publication or Scribe nodes. Collective assembly owns integration, selected review gates, merge and final recording. The inspected collective path does not establish automatic PR publication. The separate built-in default explicitly contains a publish/reuse-PR action; that action is not proof that commits were pushed or publication succeeded.

CatalogWorkflowBindingTests checks catalog gates and a bindability subset. Its bindability theory omits agent-evaluation; that is a test-coverage distinction, not evidence of unsupported fan-out nodes. The current evaluation graph uses prompt steps.

software-delivery ​

Plan and implement, pass QA and RAI, receive rubberduck feedback and a code-review prompt, then pass Build/Test and human review. Revision branches return to implementation; decline and safety failure have explicit terminals.

bug-fix ​

Triage and fix, then QA peer verification, RAI, Build/Test and human review. The lighter intake does not remove safety or build gates.

content-authoring ​

Research, draft and edit are prompt steps. RAI precedes human review; approval proceeds to a publish prompt, not a merge executor. No-change and failure branches remain visible.

pm-discovery ​

Research, synthesis and stakeholder review are prompts, followed by a distinct human gate. Requested changes return to synthesis; approved work reaches the authored Done terminal.

agent-evaluation ​

Evaluation setup, runs and collection are sequential prompt nodes, not fan_out/fan_in. The safety gate can request revision, stop unsafe work, finish without changes, or allow the report prompt.

incident-response ​

Triage, mitigation and verification are prompts. Human approval leads to the postmortem prompt; requested changes return to mitigation and decline terminates without a postmortem.

infra-ops ​

Plan and implement, then DevOps peer validation (pass/fail), RAI, infrastructure peer review and human review (approved/request-changes/declined).

Blueprint workflow mappings ​

BlueprintDeclared workflow set, in source order
Software Developmentsoftware-delivery, bug-fix
Product Managementpm-discovery, content-authoring
Content Authoringcontent-authoring
Product & Software Deliverypm-discovery, software-delivery, bug-fix
AI Agent Engineeringagent-evaluation, software-delivery, bug-fix

These arrays come from the five embedded blueprint JSON files. Source order is not a guarantee that automatic selection always chooses the first entry. Incident response and infrastructure operations remain catalog workflows without a dedicated blueprint in that embedded set.

The built-in default supplies Agent -> RAI -> human Review -> Merge -> publish/reuse PR -> Scribe, with explicit no-change, revision, blocked and failure branches. See workflow binding and selection.

Diagram details and constraints
ElementContract
titleAgent Evaluation workflow
subtitleAuthored graph • all YAML branches retained
returns-headingSOURCE / RETURN
outcomes-headingOUTCOMES
footerEvaluation Runs and Collect Results are prompt steps, not fan-out / fan-in.
Evaluation SetupEvaluation Setup
Evaluation SetupAgent task
Evaluation Setupagent
Evaluation RunsEvaluation Runs
Collect ResultsCollect Results
Safety GateSafety Gate
Safety GateVerdict routing
Safety Gaterai
Evaluation ReportEvaluation Report
Safety FailedSafety Failed
Safety FailedWorkflow endpoint
DoneDone
edge-04-labelrevise
edge-05-labelsafety- failed
edge-06-labelno- changes
edge-07-labelreview
sequential-prompts-titleSequential evaluation
sequential-prompts-subSetup, runs and collection are prompt tasks. No parallel split or join is declared.
sequential-prompts-metaagent_evaluation.yaml · nodes
Diagram details and constraints
ElementContract
titleBug Fix workflow
subtitleAuthored graph • all YAML branches retained
returns-headingSOURCE / RETURN
outcomes-headingOUTCOMES
footerSolid: advance / outcome Dashed marigold: revision / return
TriageTriage
TriageAgent task
Triageagent
FixFix
VerifyVerify
VerifyIndependent peer review
Verifyqa-engineer
RAI CheckRAI Check
RAI CheckVerdict routing
RAI Checkrai
Build & TestBuild & Test
Build & TestBuild and test verification
Human ReviewHuman Review
Human Reviewhuman-review
Safety FailedSafety Failed
Safety FailedWorkflow endpoint
DeclinedDeclined
DoneDone
edge-03-labelapproved
edge-04-labelrequest-changes
edge-05-labeldeclined
edge-06-labelrevise
edge-07-labelsafety- failed
edge-08-labelno- changes
edge-09-labelreview
Diagram details and constraints
ElementContract
titleContent Authoring workflow
subtitleAuthored graph • all YAML branches retained
returns-headingSOURCE / RETURN
outcomes-headingOUTCOMES
footerSolid: advance / outcome Dashed marigold: revision / return
ResearchResearch
ResearchAgent task
Researchagent
DraftDraft
Editorial ReviewEditorial Review
Editorial Reviewreview
RAI CheckRAI Check
RAI CheckVerdict routing
RAI Checkrai
Human ReviewHuman Review
Human Reviewhuman-review
PublishPublish
Safety FailedSafety Failed
Safety FailedWorkflow endpoint
DeclinedDeclined
DoneDone
edge-04-labelrevise
edge-05-labelsafety- failed
edge-06-labelno- changes
edge-08-labelapproved
edge-09-labelrequest-changes
edge-10-labeldeclined
Diagram details and constraints
ElementContract
titleIncident Response workflow
subtitleAuthored graph • all YAML branches retained
returns-headingSOURCE / RETURN
outcomes-headingOUTCOMES
footerSolid: advance / outcome Dashed marigold: revision / return
TriageTriage
TriageAgent task
Triageagent
MitigateMitigate
VerifyVerify
Verifyreview
Review GateReview Gate
Review GateVerdict routing
Review Gatehuman-review
PostmortemPostmortem
DeclinedDeclined
DeclinedWorkflow endpoint
DoneDone
edge-04-labelapproved
edge-05-labelrequest-changes
edge-06-labeldeclined
VerifyVerify is a prompt
VerifyThe review role prepares evidence. Review Gate owns the approval, revision and decline routes.
Verifyincident_response.yaml · nodes
Diagram details and constraints
ElementContract
titleInfrastructure & Operations workflow
subtitleAuthored graph • all YAML branches retained
returns-headingSOURCE / RETURN
outcomes-headingOUTCOMES
footerSolid: advance / outcome Dashed marigold: revision / return
PlanPlan
PlanAgent task
Planagent
ImplementImplement
ValidateValidate
ValidateIndependent peer review
Validatedevops-engineer
RAI CheckRAI Check
RAI CheckVerdict routing
RAI Checkrai
Infra & Config ReviewInfra & Config Review
Infra & Config Reviewsecurity-engineer
Human ReviewHuman Review
Human Reviewhuman-review
Safety FailedSafety Failed
Safety FailedWorkflow endpoint
DeclinedDeclined
DoneDone
edge-03-labelpass
edge-04-labelfail
edge-05-labelrevise
edge-06-labelsafety- failed
edge-07-labelno- changes
edge-08-labelreview
edge-09-labelapproved
edge-10-labelrequest-changes
edge-11-labeldeclined
Diagram details and constraints
ElementContract
titleProduct discovery workflow
subtitleAuthored graph • all YAML branches retained
returns-headingSOURCE / RETURN
outcomes-headingOUTCOMES
footerSolid: advance / outcome Dashed marigold: revision / return
ResearchResearch
ResearchAgent task
Researchagent
SynthesisSynthesis
Stakeholder ReviewStakeholder Review
Stakeholder Reviewreview
Review GateReview Gate
Review GateVerdict routing
Review Gatehuman-review
DeclinedDeclined
DeclinedWorkflow endpoint
DoneDone
edge-04-labelapproved
edge-05-labelrequest-changes
edge-06-labeldeclined
discovery-output-titleDiscovery output
discovery-output-subResearch, requirements and feature definition produce documents and specs, not deployable code.
discovery-output-metapm_discovery.yaml · description
stakeholder-contract-titlePrepare, then decide
stakeholder-contract-subStakeholder Review prepares synthesis for approval. Only Review Gate emits verdicts.
stakeholder-contract-metareview prompt · human-review gate
Diagram details and constraints
ElementContract
titleSoftware Delivery workflow
subtitleAuthored graph • all YAML branches retained
returns-headingSOURCE / RETURN
outcomes-headingOUTCOMES
footerSolid: advance / outcome Dashed marigold: revision / return
PlanPlan
PlanAgent task
Planagent
ImplementImplement
Test GateTest Gate
Test GateIndependent peer review
Test Gateqa-engineer
RAI CheckRAI Check
RAI CheckVerdict routing
RAI Checkrai
Rubberduck ReviewRubberduck Review
Rubberduck Reviewrubberduck
Code ReviewCode Review
Code Reviewreview
Build & TestBuild & Test
Build & TestBuild and test verification
Review GateReview Gate
Review Gatehuman-review
Safety FailedSafety Failed
Safety FailedWorkflow endpoint
DeclinedDeclined
DoneDone
edge-03-labelpass
edge-04-labelfail
edge-05-labelrevise
edge-06-labelsafety- failed
edge-07-labelno- changes
edge-12-labelapproved
edge-13-labelrequest-changes
edge-14-labeldeclined

Visual model ​

Default workflow ​

Flowchart of the six-stage default workflow: agent production, Responsible AI gate, human review, merge attempt, pull-request publication attempt, and Scribe recording, including revision, no-change, decline, safety, and blocked-merge paths.

Structured source · Editable draw.io