Conclave

Conclave — Scrum for Claude Code and Cursor Teams

Conclave is the methodology layer this plugin implements: Scrum, executed by a distributed engineering team where every team member uses Claude Code or Cursor locally, and the shared state is plain markdown committed to git.

This repository ships two installable packages (see ADR-002): the Claude Code plugin at the repo root (conclave) and the Cursor plugin under platforms/cursor/ (conclave-cursor). Both read and write the same target-repo conclave/ contract. This SKILL.md is the canonical methodology source; the Cursor package receives a synced copy via scripts/sync-cursor-platform.sh.

This skill documents:

1. The Scrum model Conclave assumes
2. The directory layout Conclave reads and writes
3. The role-to-subagent mapping
4. How slash commands invoke role subagents

The slash commands (/conclave-init, /conclave-planning, /conclave-standup, etc.) consume this skill for context. Role charters under agents/ are loaded by name from those slash commands (Claude Code: skills/conclave/agents/; Cursor: platforms/cursor/agents/).

---

1. The Scrum model Conclave assumes

Conclave assumes a standard Scrum setup with a small accommodation for real engineering teams:

Scrum concept	Conclave term	Notes
Development Team	Disciplines: Tech Lead, Frontend, Backend, QA, Designer, DevOps	Always present in the roster, whether or not they map to six different people (v0.2.0+). This is the primary roster axis — see conclave/team/roster.md's Discipline column.
Product Owner	Product Manager (PM)	An optional process role (v0.2.0+), not a discipline — any discipline-holder can additionally carry it. Same responsibilities when someone does (own the backlog, prioritize, define acceptance). We call it PM because most teams in practice do.
Scrum Master	Scrum Master (SM)	An optional process role (v0.2.0+), not a discipline. Facilitates ceremonies, removes blockers, when someone holds it. If nobody does, the Tech Lead and team decide process by consensus.
Product Backlog	conclave/product/backlog.md	Ordered list of user stories.
Sprint Backlog	conclave/sprints/SPRINT-NNN/spec.md selected stories table	Snapshot at planning time.
Increment	The merged PRs that close stories	Conclave does not track this directly; git does.
Sprint Planning	/conclave-planning (out of MVP scope)	Locks the sprint.
Daily Scrum	/conclave-standup per dev (out of MVP scope)	Logs into sprints/SPRINT-NNN/daily/.
Sprint Review	/conclave-review (out of MVP scope)	Demo + acceptance.
Sprint Retrospective	/conclave-retro (out of MVP scope)	What to keep, change, start.
Definition of Ready	conclave/product/definition-of-ready.md	Team-customized checklist.
Definition of Done	conclave/product/definition-of-done.md	Team-customized checklist.
User story	One file under sprints/SPRINT-NNN/stories/	INVEST format.
Acceptance criteria	One file under sprints/SPRINT-NNN/acceptance/	Gherkin Given/When/Then.

---

2. Directory layout Conclave reads and writes

At the root of the team's repo:

conclave/                             # VISIBLE top-level directory, all markdown
├── README.md                         # explains the directory to anyone browsing on GitHub
├── config.md                         # project type, stack, paths, project_language (frontmatter + prose)
├── team/
│   ├── roster.md                     # team members, discipline(s), optional PM/SM process role(s)
│   ├── ceremonies.md                 # sprint length, planning day, standup time, retro day
│   ├── testing-environments.md       # CI env-var/secret NAMES the generated UAT tests read — never real values
│   ├── board.md                      # branding for conclave-board/ (company name, logo, colors) — no secrets
│   └── PR_REVIEW_TEMPLATE.md         # PR review checklist template for team use (written by /conclave-init)
├── product/                          # persists across sprints
│   ├── backlog.md                    # ordered Product Backlog
│   ├── architecture.md               # living architectural doc (ADRs)
│   ├── definition-of-ready.md        # team-agreed DoR
│   ├── definition-of-done.md         # team-agreed DoD
│   └── bugs/                         # BUG-NNN-<slug>.md via /conclave-bug report — flat, no index
├── context/                          # frozen snapshots of inputs used (auditable)
│   ├── claude-md.snapshot.md
│   ├── skills.inventory.md
│   └── rules.inventory.md
├── report/                           # sprint closing reports and DORA data (v0.16.0+)
│   ├── SPRINT-NNN/
│   │   ├── report.md                 # sprint closing report (written by /conclave-sprint at close)
│   │   ├── UAT.md                    # functional UAT guide for the sprint (written by /conclave-sprint)
│   │   └── dora-data.yml             # DORA raw data snapshot for /conclave-dora to aggregate
│   └── dora/
│       └── DORA-NNN-<period>-<date>.md  # generated by /conclave-dora
├── runs/                             # delivery-loop run reports, only in repos with no sprint at all
│   └── RUN-NNN-dev-loop.md           # written by /conclave-dev --loop on a bug-only repo
└── sprints/
    └── SPRINT-NNN/
        ├── meta.md                   # name, dates, goal, status
        ├── spec.md                   # sprint plan
        ├── planning.md               # planning ceremony record
        ├── stories/
        │   └── US-NNN-<slug>.md
        ├── acceptance/
        │   └── AC-US-NNN.md
        ├── bugs/                     # QA-detected bugs (v0.16.0+) — BUG-NNN-<slug>.md
        │                             # linked to story + acceptance + PR that introduced them
        │                             # Critical bugs block sprint close
        └── runs/                     # delivery-loop run reports
            └── RUN-NNN-dev-loop.md   # /conclave-dev --loop (v0.15.0+; RUN-NNN-autonomous-loop.md
                                      # files from 0.13.0/0.14.0 stay on disk, never rewritten)

GitHub templates written by /conclave-init (outside conclave/, not part of this contract):

.github/
├── ISSUE_TEMPLATE/
│   └── bug_report.md                 # GitHub bug report form (Conclave-linked)
└── PULL_REQUEST_TEMPLATE.md          # GitHub PR template (Conclave-linked)

Invariants every Conclave command must respect

• Markdown only. Structured data lives in YAML frontmatter at the top of each file. The body below is human-readable prose. No JSON-only files, no SQLite, no binaries.
• Visible directory. conclave/ is committed and renders on GitHub.
• Append, do not clobber. A second /conclave-planning run on a new sprint creates SPRINT-002/, not overwriting SPRINT-001/. The backlog is updated additively.
• Snapshot context. Every artifact-generating command writes a fresh snapshot under conclave/context/ so the artifact is auditable against the inputs that produced it.
• Reference, don't duplicate. Stories reference their acceptance file (See acceptance/AC-<PREFIX>-NNN.md); sprint spec references product/definition-of-done.md rather than copying it.
• Numbering is sticky. SPRINT-NNN and <story_prefix>-NNN IDs increment monotonically and are never reused.
• story_prefix governs story IDs (v1.1.0+). The story_prefix field in config.md (default US) is the prefix for all story and acceptance file names: US-001-slug.md / AC-US-001.md, or TASK-001-slug.md / AC-TASK-001.md if overridden. Set once by /conclave-init; hand-edit if the team decides to change it (existing files are not renamed).
• product_doc_path is the planning source of truth (v1.1.0+). The product_doc_path field in config.md points to the product planning document (e.g. docs/mvp.md). /conclave-planning reads this file to generate the backlog, architecture, stories, and acceptance criteria. It is set by /conclave-init and can be updated to point to a different document at any time.
• Roster schema degrades gracefully. A roster.md written before v0.2.0 (no Discipline column) is not rejected — commands that read it treat every member as multi-discipline and print a one-time compatibility hint. No auto-migration is provided; a team opts into discipline-based assignment by re-running /conclave-init or hand-editing the roster.
• UAT config degrades gracefully. A testing-environments.md that doesn't exist yet, or still has every row TBD (v0.2.0 installs, or a fresh /conclave-init before the team fills it in), is not a hard failure — /conclave-qa skips UAT generation entirely and verifies acceptance criteria exactly as it did before v0.3.0.
• conclave-board/ (v0.5.0+) is application code, not part of this contract. /conclave-board scaffolds a Next.js app as a sibling of conclave/, not inside it — the markdown-only invariant above applies only to conclave/ itself. The board reads conclave/ but never writes to it.
• Bugs (v0.10.0+) skip Sprint Planning by design. A BUG-NNN reported via /conclave-bug report is written directly in status: ready under conclave/product/bugs/ — not under any sprints/SPRINT-NNN/. /conclave-planning and /conclave-sprint never look inside conclave/product/bugs/; a bug is picked up directly via /conclave-dev BUG-NNN, and driven all the way to an approved PR via /conclave-dev --loop BUG-NNN (v0.15.0+).
• QA-detected bugs (v0.16.0+) go in the sprint's own bugs folder. When /conclave-qa finds a blocking defect during verification on the integration branch, it writes BUG-NNN-<slug>.md to conclave/sprints/SPRINT-NNN/bugs/ (not conclave/product/bugs/). These bugs are linked to the story, acceptance criteria, and the PR that introduced the regression. Critical sprint bugs block /conclave-sprint from generating the closing report — the sprint cannot close until they are resolved.
• project_language governs all generated prose (v0.16.0+). The project_language field in config.md (ISO 639-1 code, default es) is read by every command that generates human-readable markdown. Role subagents receive it as an explicit instruction to write stories, acceptance criteria, reports, bug descriptions, and comments in that language. Stack/code identifiers, DORA metric names, and frontmatter keys remain in English.
• Sprint close reports and UAT guides (v0.16.0+) live under conclave/report/SPRINT-NNN/. They are generated by /conclave-sprint when the sprint closes, after the critical-bug gate passes. The DORA data snapshot (dora-data.yml) is also written there for /conclave-dora to aggregate.
• Multi-sprint planning (v0.16.0+, redesigned v1.1.0+). /conclave-planning --all reads the product document, generates stories for all sprints, and plans every sprint in one pass. The first sprint becomes active; the rest remain draft with their planning records written, ready to be activated when the prior sprint closes.
• Run reports are append-only and double as locks. runs/RUN-NNN-*.md files are never deleted or rewritten by a later run; RUN-NNN increments monotonically within its directory. A report with outcome: in_progress blocks a second run whose scope overlaps it. conclave/runs/ exists only as the fallback home for a dev-loop report in a repo that has no sprints/ at all (bug-only work) — when any sprint exists, reports live under that sprint.
• No command merges a pull request. Since v0.15.0 (ADR-006) nothing in Conclave runs gh pr merge. QA verification and Tech Lead approval are gates; landing the code is a human action.

---

3. Role-to-subagent mapping

Role charters are markdown files under skills/conclave/agents/. They have no frontmatter — they are pure prose loaded by slash commands when delegating work.

Subagent file	Used by (shipped)	Used by (planned)
agents/product-manager.md	/conclave-planning (Phase A: backlog generation from product doc; Phase B: scope review Wave 1), /conclave-story (new / edit / split — retire is mechanical and skips this agent)	/conclave-groom, /conclave-review
agents/tech-lead.md	/conclave-planning (Phase A: architectural foundation from product doc; Phase B: feasibility review + discipline assignment, Wave 1), /conclave-pr-review (code review + approval), /conclave-adr (topic-directed and discovery ADR authoring)	/conclave-substack
agents/scrum-master.md	/conclave-planning (facilitator + assignment, Wave 2 — runs after PM/TL)	/conclave-standup, /conclave-review, /conclave-retro
agents/developer.md	/conclave-dev US-NNN|BUG-NNN [US-NNN|BUG-NNN ...] (items with discipline: frontend | backend | mobile | multi, or unset) — one Agent call per item, ≤ 3 concurrent per batch, story and bug IDs may be mixed in one invocation. For a BUG-NNN, reproduces via the bug file's inline Gherkin repro steps before fixing, and the rendered PR body includes Fixes #<github_issue_number> (v0.10.0+). Autonomous mode (v0.9.0+): --no-interaction CLI flag or commands.dev.interactive: false in config.md makes the command run headless — no AskUserQuestion prompts; defaults or AUTONOMOUS_ABORT: <reason>; per-run report appended to the file; ends at review, never merges. /conclave-sprint Phase 2 always forces autonomous (stories only — see below). Autonomous Three-Wave Delivery Loop (v0.15.0+): --loop or commands.dev.loop: true takes the active sprint (or the IDs passed) and runs W1 Dev + green CI → W2 QA → W3 forced TL review, with any wave failure returning the affected stories to W1; W0 orders the scope by dependencies: and serializes file overlaps. Recurring local-time schedule + budgets from commands.dev.*, run report RUN-NNN-dev-loop.md with token and agent-productivity statistics, Slack templates. Implies autonomous; accepts BUG-NNN; never merges; never closes a sprint. See ADR-006.	—
agents/designer.md	/conclave-dev US-NNN [US-NNN ...] (stories with discipline: design)	—
agents/devops.md	/conclave-dev US-NNN [US-NNN ...] (stories with discipline: devops)	—
agents/qa.md	/conclave-qa US-NNN|BUG-NNN [US-NNN|BUG-NNN ...] — one Agent call per item, ≤ 3 concurrent per batch, story and bug IDs may be mixed. A bug's repro steps are verified exactly like a story's Gherkin scenarios.	—
agents/qa.md (again)	/conclave-bug report (v0.10.0+) — one Agent call per invocation, authors Gherkin repro steps + an advisory severity note from the report's raw input. /conclave-bug list is mechanical (frontmatter-only) and skips the agent, same precedent as /conclave-story retire.	—
(all of the above)	/conclave-sprint — sequential four-phase one-pass runner (Planning → Dev batch-of-3 → QA batch-of-3 → PR review if peer_pr_review.required). Headless one-pass (--no-interaction / commands.sprint.interactive: false) is the same pass with documented planning defaults and zero prompts. Neither mode merges, self-heals, reads a schedule, or spends a budget — since v0.15.0 unattended delivery is /conclave-dev --loop (ADR-006). Each Agent/Task call uses the role model from models:.	—
agents/product-manager.md (again)	/conclave-story <new|edit|split> — one Agent call per invocation. /conclave-story retire is mechanical (frontmatter-only) and skips the agent. Available in every team_mode (solo, lean, full-scrum).	—
agents/tech-lead.md (again)	/conclave-adr [topic] — topic-directed mode writes a full ADR to conclave/product/adr/ADR-NNN-<slug>.md; discovery mode (no args) proposes 1–3 candidates then authors the picked one. Migrates any pre-0.8.0 inline ADRs in architecture.md on first run (per-ADR atomic, resumable, idempotent). Available in every team_mode.	—
agents/tech-lead.md or agents/product-manager.md	/conclave-dora [--period <type>] [--from <date>] [--to <date>] (v0.16.0+) — generates a DORA metrics report aggregating sprint close data from conclave/report/. Uses TL for full-scrum profiles (engineering-depth analysis), PM for lean/solo profiles (product-centric insights). Lean/solo output omits individual contributor breakdown.	—

Model configuration (v0.7.0+): commands read an optional models: block from conclave/config.md frontmatter. Resolution per Agent call: models.overrides.<role> → models.default → parent session model (silent no-op when block is absent). Invalid model name → warn once and fall back. Role keys: product_manager, tech_lead, scrum_master, developer, designer, devops, qa.

A slash command delegates by spawning an Agent subagent and passing the full content of the role charter file as the system prompt prefix, followed by the task-specific instructions and the context the role needs.

Multi-story concurrency: When /conclave-dev or /conclave-qa is invoked with multiple US-NNN/BUG-NNN arguments (v0.10.0+ accepts either kind, mixed freely), the orchestrator validates all items upfront (direct file reads — no Agent calls), partitions them into batches of ≤ 3, and issues all Agent calls within a batch in a single message so they run concurrently. Failures are isolated per item: a failed item is reset to ready (dev) or left at its last known state (QA) and reported in the final summary table; other items in the batch continue unaffected. /conclave-pr-review (single-ID only, no batching) also accepts either US-NNN or BUG-NNN.

---

4. How slash commands invoke role subagents

The orchestration pattern is the same one code-review uses: prose instructions inside the slash command's markdown body. There is no DSL.

• Claude Code: when the body says "Spawn a subagent loaded with skills/conclave/agents/tech-lead.md…", Claude reads the role charter, dispatches an Agent tool call with that content as context, and continues when the subagent returns. Two role subagents can run in parallel by issuing both Agent tool calls in a single message.
• Cursor (platforms/cursor/): the same ceremony steps use the Task tool (or Cursor custom agents under agents/<role>.md) and prefer AskQuestion for structured prompts in top-level Agent chat. Methodology and templates are identical (synced from this canonical tree).

---

5. Templates

All Conclave-managed artifacts are produced by filling in templates from skills/conclave/templates/. The orchestrator reads the template, replaces {{placeholders}}, and writes the resulting markdown to the team's conclave/ directory.

Templates available:

• conclave-readme.template.md
• config.template.md
• roster.template.md
• ceremonies.template.md
• definition-of-ready.template.md
• definition-of-done.template.md
• product-backlog.template.md
• architecture.template.md
• sprint-meta.template.md
• sprint-spec.template.md
• story.template.md
• acceptance.template.md
• planning.template.md
• pr-body.template.md
• verification-report.template.md
• testing-environments.template.md
• uat-report.template.md
• board.template.md
• adr.template.md
• autonomous-run.template.md
• bug.template.md
• sprint-run-report.template.md — filled by /conclave-dev --loop (mode: autonomous-dev-three-wave, scope: sprint or the invoked IDs); carries the token ledger, agent-productivity table, conflicts, and the PRs awaiting a human merge
• slack-loop-success.template.json — posted when the loop completes with everything approved
• slack-loop-partial.template.json — posted when the loop finishes with stories incomplete or drained on budget/schedule
• slack-loop-hitl.template.json — posted the moment a blocker needs a human (structural abort, dependency cycle, missing gh, attempts exhausted, pending_uat)

---

6. What is mandatory vs skippable

Conclave separates structural invariants (you cannot do Scrum without them) from ceremonies (process gates the team chooses to commit to).

Always required (structural — never skippable)

• A Sprint Plan. Without a goal and a locked story list, there is no sprint. Enforced by /conclave-planning and the existence of conclave/sprints/SPRINT-NNN/spec.md.
• Acceptance criteria on every story. Every story file must reference a non-empty acceptance/AC-US-NNN.md with Gherkin scenarios. Stories without them fail the DoR.
• QA verification of acceptance criteria. Every done story carries a verification report appended to its acceptance file. Without this, done means nothing. Enforced by /conclave-qa.
• Definition of Done compliance. The team-customized DoD checklist must be met for every story. The structural items of the DoD are non-negotiable; some items become conditional (see below).

Who approves the PR (two gates, two roles)

Conclave separates two distinct checks on a finished story:

1. QA verification — does the implementation match the acceptance criteria behaviorally? Owned by the QA role, run via /conclave-qa US-NNN. Always required.
2. Tech Lead PR approval — does the code meet the architecture, ADRs, and code-level DoD items? Owned by the Tech Lead role, run via /conclave-pr-review US-NNN. Required only when ceremonies.peer_pr_review.required: true.

The two gates do NOT collapse. QA never runs gh pr review --approve. The TL does. When the flag is off (typical for lean), QA's pass implicitly approves the PR because there is no separate technical gate — except in the three-wave delivery loop (/conclave-dev --loop, v0.15.0+, ADR-006), which ephemerally forces TL review for the run without rewriting config.md.

Neither gate authorizes a merge. Since v0.15.0 no Conclave command runs gh pr merge: the loop's terminal state is an approved PR listed in the run report, and a human lands it.

The three-wave delivery loop: waves, scheduling, budgets (v0.15.0+)

/conclave-dev --loop (or commands.dev.loop: true) is the only autonomous delivery loop. It runs W0 conflict ordering, W1 Dev to green CI, W2 headless QA, W3 forced Tech Lead review, and returns any failing story to W1 — a Tech Lead asking for changes invalidates the QA verdict that preceded it, so QA always re-runs after Dev. Waves never overlap; batch-of-3 concurrency exists only inside W1 for stories with no dependency or file overlap between them.

commands.dev.schedule is a recurring local-time gate: timezone (IANA), days, start_time / end_time (may cross midnight), duration_days, active_from, enforce. Outside the window the command no-ops with no writes at all. The pre-0.15.0 window_start / window_end pair is refused with a migration message rather than reinterpreted. commands.dev.budgets caps attempts, CI wait, wall-clock, and a best-effort token ledger; the report discloses whether token totals are estimated, measured, or mixed.

Conclave still ships no scheduler — an external trigger (Claude /loop / /schedule, a Cursor Automation, cron) invokes the command and the gate decides whether it does anything. Budget and window aborts still finalize the run report at SPRINT-NNN/runs/RUN-NNN-dev-loop.md (or conclave/runs/ in a repo with no sprint). A report with outcome: in_progress is the concurrency lock; two loops with non-overlapping scopes may run at once.

Slack is optional and template-driven: success, partial, and human-in-the-loop. HITL alerts are posted at the moment the blocker occurs so the operator can act while the run continues elsewhere. Only the env var name lives in config; a delivery failure never fails the run.

The loop requires the GitHub CLI (gh) installed and authenticated with repo access (push, PRs, review). Conclave declares the tools it uses but does not install or configure gh, and does not manage repository permissions.

Story status transitions, profile-aware

backlog → ready → in-progress → review → [verified] → done
                                              ↘
                                                retired  (parallel terminal — via /conclave-story)

• review → verified: only happens when peer_pr_review.required: true. QA pass moves the story here while waiting for TL approval.
• review → done: direct, when peer_pr_review.required: false. QA pass and PR approval collapse into the same step.
• verified → done: TL approves the PR via /conclave-pr-review.
• Any failure: back to review. The dev fixes, pushes, then QA re-verifies (and TL re-reviews if applicable).
• retired (v0.8.0+) — a parallel terminal state to done. Entered via /conclave-story retire (explicit retirement with retirement_reason and retired_at set) or /conclave-story split (on the parent, when it is decomposed into children — superseded_by: also populated). A retired story is excluded from every command's story collection (/conclave-planning, /conclave-dev, /conclave-qa, /conclave-pr-review, /conclave-sprint) — it is a historical record only. There is no un-retire command; teams that change their mind hand-edit the frontmatter (git preserves the audit trail). /conclave-planning (Phase A) is intentionally exempt from the filter — it authors new stories rather than collecting existing ones.
• UAT pending (v0.3.0+, no new status value). When testing-environments.md is configured, /conclave-qa generates CI-runnable UAT tests (Playwright/Newman for frontend/backend/multi, a manual checklist for mobile) and folds the result into its verdict. A mobile story whose checklist is awaiting or mid-completion produces verdict: pending_uat — the story frontmatter stays review, same as a real failure, but the appended section is ## QA pending, not ## QA blockers, since nothing has actually failed yet. A failed CI run on the generated tests is treated exactly like a failing Gherkin scenario.
• BUG-NNN artifacts (v0.10.0+) reuse this exact state machine. A bug reported via /conclave-bug report is written directly in status: ready — bugs never pass through backlog or Sprint Planning; /conclave-bug report is the only way one is created, and it always starts dev-ready. From there it follows the identical path a story does (/conclave-dev → /conclave-qa → /conclave-pr-review if applicable), including retired as a manual escape hatch (no /conclave-bug retire sub-action yet — hand-edit the frontmatter).

Skippable per team profile

The team chooses a profile in conclave/config.md (team_profile: lean | full-scrum | custom) and Conclave's ceremony commands read it. Skipped ceremonies are silently a no-op; required ceremonies are enforced.

Ceremony	Command	lean default	full-scrum default	Notes
Daily Standup	/conclave-standup	off	on	Logs to sprints/SPRINT-NNN/daily/.
Backlog Grooming	/conclave-groom	off	on	When off, grooming happens inside /conclave-planning.
Peer PR Review	(DoD check)	off	on	Solo devs and small teams often skip this. The Dev agent still self-reviews.
Sprint Review	/conclave-review	off	on	Required when there are stakeholders to demo to.
Sprint Retrospective	/conclave-retro	off	on	First thing to get dropped under pressure; team should opt back in when it stabilizes.

Profile semantics

• lean — only the structural invariants are enforced. Intended for solo devs, very small teams (2–3), and internal/tooling work.
• full-scrum — every ceremony is required. Intended for cross-functional teams that ship to external stakeholders.
• custom — the team sets each ceremonies.*.required flag individually. The profile is recorded as custom so it is obvious nobody is following a preset.

How commands respect the profile

When a future ceremony command runs (e.g. /conclave-standup), the first thing it does is read conclave/config.md and check its required: flag.

• If required: true, the command runs normally.
• If required: false and the user invoked the command explicitly, it still runs but prints a hint that it is optional in this profile.
• If required: false and the command is triggered indirectly (e.g. as a step inside /conclave-close-sprint), it is skipped silently.

The two always-required gates (sprint_planning, qa_verification) cannot be flagged off — attempting to set required: false for them is rejected with a clear error.

---

7. Visual boards

Conclave ships two complementary boards. Neither replaces the other; neither writes story/sprint source-of-truth under conclave/ (except /conclave-board may create team/board.md once).

7.1 Status Kanban — /conclave-board (v0.5.0+)

/conclave-board is not a prose-orchestrated subagent — it's a one-time scaffold plus a deterministic background sync, with no Agent/Task call anywhere in its update loop:

• Scaffold, once: copies a Next.js + shadcn/ui boilerplate into conclave-board/ (a sibling of conclave/) and renders conclave/team/board.md for branding. A second run refuses, same idempotency posture as /conclave-init.
• Stay current, automatically: a PostToolUse / Cursor afterFileEdit hook re-runs conclave-board/scripts/generate-data.mjs when paths under conclave/ change and a board is scaffolded.
• Read-only toward stories: status changes still only happen through /conclave-dev, /conclave-qa, and /conclave-pr-review.
• Local only: no CI pipeline, no hosting, no cross-machine sync.

See docs/specs/conclave-board/spec.md.

Glossary

• Founding artifacts. The minimum set a team needs to start working in Scrum: roster, ceremonies, DoR, DoD, Product Backlog, Architectural Foundation, and Sprint 1 plan. Conclave's MVP produces all of these.
• Sprint spec. The locked plan for one sprint: goal + selected stories + reference to DoD. Lives at conclave/sprints/SPRINT-NNN/spec.md.
• Context snapshot. A point-in-time copy of CLAUDE.md, available skills, and detected rules, written to conclave/context/ whenever an artifact-generating command runs.
• Setup wizard. /conclave-init — one-time project initialization wizard (no AI agents). Collects project name, story prefix, stack, launch date, and product document path. Creates the workspace. (/conclave-spec is a deprecated alias that redirects here.)
• Sprint planner. /conclave-planning — reads the product document and generates the full backlog, architectural foundation, stories, and acceptance criteria (Phase A, first run only), then runs the planning ceremony and activates the sprint (Phase B, every run). Use --all to plan every sprint from the document at once.