agentsclimarketplace

Roundtable execution

Skill spec2ship/spec2ship/skills/roundtable-execution

This skill provides instructions for executing multi-agent roundtable discussions. Use when a command needs to run discussion rounds with facilitator and participants. Referenced by: roundtable.md (master orchestrator); specs.md, design.md, brainstorm.md (thin launchers that Read-and-follow roundtable.md). Trigger: 'execute roundtable', 'run discussion rounds', 'multi-agent discussion'.From its SKILL.md

Install
npx -y skills add spec2ship/spec2ship --skill roundtable-execution

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

11.2 KB, as published. Nobody here has run it

Roundtable Execution Skill

This skill provides the canonical, profile-aware reference for executing multi-agent roundtable discussions in spec2ship. After TECH-002 Phase 7B (2026-05), the Phase 2 Round Execution Loop is extracted to references/phase-2-core.md as a single executable source. After TECH-002 Phase 8 (2026-05), commands/roundtable.md is the master orchestrator for all 4 workflow types (specs, design, brainstorm, roundtable); specs.md, design.md, brainstorm.md are thin launchers that Read-and-follow the master.

When to Use This Skill

  • Executing /s2s:specs requirements gathering (thin launcher delegates to master)
  • Executing /s2s:design architecture design (thin launcher delegates to master)
  • Executing /s2s:brainstorm ideation sessions (thin launcher delegates to master)
  • Executing /s2s:roundtable (master, native or with --workflow-type)

Workflow Context

Each workflow has specific goals, participants, artifacts, and outputs. Authoritative source: the YAML profile in profiles/{workflow}.yaml — the tables below are a human-readable summary.

specs Workflow

AspectValue
GoalDefine WHAT to build — requirements, constraints, scope
Default Participantsproduct-manager, ux-researcher, business-analyst, qa-lead
Default Strategyconsensus-driven
Primary ArtifactsREQ-*, BR-*, NFR-*
Secondary ArtifactsOQ-*, CONF-*, EX-*
Output.s2s/requirements.md
Profileprofiles/specs.yaml
Agendareferences/agenda-specs.md

design Workflow

AspectValue
GoalDefine HOW to build — architecture, components, interfaces
Default Participantssoftware-architect, security-champion, technical-lead, devops-engineer
Default Strategydebate
Primary ArtifactsARCH-*, COMP-*, INT-*
Secondary ArtifactsADR-*, OQ-*, CONF-*
Output.s2s/architecture.md + .s2s/decisions/
Profileprofiles/design.yaml
Agendareferences/agenda-design.md

brainstorm Workflow

AspectValue
GoalExplore possibilities — ideas, risks, mitigations
Default Participantsproduct-manager, software-architect, technical-lead, devops-engineer (configurable via --participants)
Default Strategydisney (FORCED — --strategy is ignored)
Primary ArtifactsIDEA-*, RISK-*, MIT-*
Secondary ArtifactsOQ-*, CONF-*
Output.s2s/sessions/{session-id}-summary.md + updates to .s2s/ideas.md
Profileprofiles/brainstorm.yaml
Agendareferences/agenda-brainstorm.md (phase-based, Disney)

Workflow Differences Summary

Aspectspecsdesignbrainstorm
FocusUser needs, requirementsTechnical architectureCreative exploration
ToneCollaborative agreementAdversarial evaluationNo criticism (dreamer) → Full critique (critic)
ParticipantsBusiness + QA focusTechnical focusFlexible (--participants)
StrategyConsensusDebateDisney phases (forced)
Progress axisAgenda topicsAgenda topicsDisney phase machine

Key Architecture

  • Session file: .s2s/sessions/{session-id}.yaml — slim index with embedded artifacts (per ADR-0008/0010).
  • Session folder: .s2s/sessions/{session-id}/ — subfolder for rounds/ dumps when --verbose.
  • Artifacts: EMBEDDED in session.yaml under artifacts.{session_key}, NOT separate files.
  • Verbose dumps: rounds/{NNN}-{PP}-{actor}.yaml per verbose-dump-format.md.

Phase structure

All 4 workflows (specs, design, brainstorm, native roundtable) execute the same 4 phases inside commands/roundtable.md (the master). Thin launchers (specs.md, design.md, brainstorm.md) perform workflow-specific prep (prerequisite checks, smart source detection, argument parsing) and then Read-and-follow the master. The master is profile-driven via profiles/{workflow}.yaml.

PhaseWhere it runsSource of truth
0: Auto-detect / parse argsroundtable.md PHASE 0scoped to workflow_type
1: Session setup (folder + 3 snapshots + skeleton)roundtable.md PHASE 1profiles/{workflow}.yaml
2: Round Execution Loopphase-2-core.md via roundtable.md PHASE 3 callerphase-2-core.md §2 + PROFILE
3: Completion (close, summary, output)roundtable.md PHASE 4 + output-generation/references/{workflow}.mdper-workflow output reference

Phase 1: Session Setup (master, profile-driven)

Driven by PROFILE loaded from profiles/{workflow}.yaml in roundtable.md PHASE 1:

  1. Generate session ID: {YYYYMMDD}-{workflow_type}-{topic-slug}.
  2. Create session folder: .s2s/sessions/{session-id}/[rounds/].
  3. Create snapshot files: config-snapshot.yaml, context-snapshot.yaml, and agenda.yaml if PROFILE.progress.axis == agenda.
  4. Create session index file: .s2s/sessions/{session-id}.yaml with skeleton derived from PROFILE (artifact_types populate artifacts.by_state, topics, validation; progress.axis discriminates agenda vs disney_phase).

Topic resolution per PROFILE.topic.source (cli-arg vs synthesized from project context vs none).

Phase 2: Round Execution Loop (canonical, profile-aware)

Read references/phase-2-core.md and follow its §2 (Round Loop algorithm). The algorithm loops Steps 2.0 → 2.9 internally until terminal dispatch.

Steps (per phase-2-core.md):

StepPurposeToken checkpoint
2.0Context Capacity Check—
2.1Display Round Start + update state.json—
2.2Facilitator QuestionT1
2.3Participant Responses (PARALLEL)T2
2.4Facilitator SynthesisT3
2.5Process Artifacts (uses artifact-schemas/)—
2.6Update Session File—
2.6bValidate Round Output (round-validation.md)—
2.6cDiagnostic Observation (IF --diagnostic) — MANDATORY—
2.10Phase Transition (brainstorm only, disney-phase-machine.md)—
2.7Display Round Recap—
2.8Handle Interactive Mode—
2.9Evaluate Next Action (min_rounds enforcement, dispatch)—

All step details are in phase-2-core.md. Do not duplicate Phase 2 logic in commands — commands invoke phase-2-core.md via the caller pattern documented there (§3).

Phase 3: Completion (master, output workflow-specific)

roundtable.md PHASE 4 runs the common close-out then dispatches to the workflow-specific output reference:

  1. Final Diagnostic Report (IF --diagnostic): invoke session-observer in end-session mode, display final report.
  2. Update Session Status: set status: "closed", clear state.json.active_session. Execute "Session Complete" token tracking.
  3. Read Session for Summary: extract artifacts from session file.
  4. User Review: AskUserQuestion to approve / refine / add more.
  5. Generate Output: invoke output-generation skill with the workflow-specific reference (references/specs-srs.md, references/design-arc42.md, references/brainstorm.md, or references/roundtable-summary.md). Handoff variables (OUTPUT_MERGE_MODE, OUTPUT_FORMAT, FOCUS_AREA) wired by the master per launcher context.

Output stays workflow-specific (different documents), but Phase 3 orchestration is now centralized in the master rather than duplicated per command.


Reference Files

Phase 2 canonical references (new in TECH-002 Phase 7B)

FileContent
references/phase-2-core.mdExecutable Phase 2 Round Execution Loop — single source consumed by all 3 workflow commands
references/profile-schema.mdWorkflow profile YAML schema + field-to-§1 mapping + how-to-add-a-workflow
profiles/specs.yamlspecs workflow profile (artifact types, participants, agenda axis, defaults)
profiles/design.yamldesign workflow profile
profiles/brainstorm.yamlbrainstorm workflow profile (with forced: true disney strategy and has_phase_transition: true)
references/artifact-schemas/README.mdIndex of 12 per-type artifact schemas
references/artifact-schemas/{type}.mdCanonical schema per artifact type (req, br, nfr, ex, arch, comp, int, idea, risk, mit, oq, conf)
references/disney-phase-machine.mdDisney phase state machine (brainstorm Step 2.10)
references/strategy-hooks.mdStrategy-specific Phase 2 variation hooks (debate_role, debate_phase, future hat_role): contract documented and wired via Option B (Phase 4). See references/strategy-hook-resolution.md for the 3-branch dispatch and references/strategy-resolution.md for the D3 hierarchy

Supporting references

FileContent
references/session-schema.mdFull session.yaml schema
references/agenda-specs.mdSpecs workflow agenda with DoD criteria
references/agenda-design.mdDesign workflow agenda with DoD criteria
references/agenda-brainstorm.mdBrainstorm workflow agenda (phase-based)
references/verbose-dump-format.mdVerbose dump file naming + canonical YAML schemas for facilitator/participant/synthesis/session-observer dumps
references/token-tracking.mdToken tracking script + checkpoints T1/T2/T3 + Session Complete
references/round-validation.mdPer-round validation checks (Step 2.6b)
references/diagnostic.mdDiagnostic mode notes
references/definition-of-done.mdStep validation checklist
references/error-handling.mdError recovery patterns
references/workspace-scope.mdWorkspace/component scope handling (Phase 1 Step 1.3b-1.4)

Migration history

  • v1.x–v2.x: SKILL.md inlined Phase 1/2/3 algorithms (1000+ lines), duplicating logic also present in each workflow command. Drift between SKILL.md and commands led to BUG-013 and others.
  • v3.0 (2026-05, TECH-002 Phase 7B.5): SKILL.md restructured to thin overview pointing to phase-2-core.md (executable Phase 2 single-source) and per-type artifact schemas.
  • v3.1 (2026-05, TECH-002 Phase 8): Phase structure section rewritten. Phase 1 (session setup) and Phase 3 (close-out + output dispatch) moved out of each workflow command into the master orchestrator commands/roundtable.md. Thin launchers (specs.md, design.md, brainstorm.md) Read-and-follow the master.

Referenced by: commands/roundtable.md (master orchestrator, runs PHASE 0+1+3+4 and delegates PHASE 2 to phase-2-core.md); commands/specs.md, commands/design.md, commands/brainstorm.md (thin launchers, Read-and-follow the master per Pattern 1, see ADR-0011 Phase 8 addendum).

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.