create-byan-agent 2.19.2 → 2.20.1

package/install/templates/.claude/workflows/quick-dev.js

@@ -0,0 +1,273 @@
+export const meta = {
+  name: 'quick-dev',
+  description: 'Native port of the BYAN quick-dev workflow: flexible implementation from a tech-spec OR direct instructions, with self-check, an adversarial diff review, and a structured findings verdict for the orchestrating skill to present at the human gate.',
+  phases: [
+    { title: 'MODE', detail: 'capture baseline + classify tech-spec vs direct + escalation signals (no human menu here)' },
+    { title: 'CONTEXT', detail: 'direct mode only: identify files, patterns, dependencies, infer AC + plan' },
+    { title: 'EXECUTE', detail: 'implement all tasks task-by-task; per-task convergence cap mirrors the 3-failure HALT rule' },
+    { title: 'SELF_CHECK', detail: 'audit tasks/tests/AC/patterns; update tech-spec status (mode A)' },
+    { title: 'ADVERSARIAL_REVIEW', detail: 'build the baseline diff, run adversarial review, order + number findings (zero findings is suspicious)' },
+    { title: 'VERDICT', detail: 'return findings + summary; the W/F/S resolution decision stays at the human gate' },
+  ],
+}
+
+// ---------------------------------------------------------------------------
+// FD / STRICT STATE CONTRACT  (re-asserted inline — byan-lint-workflows).
+//
+// The in-CLI Workflow tool runs this script OUTSIDE the conversation turn, so
+// BYAN's main-thread hooks (fd-phase-guard, strict-scope-guard, strict-stop-
+// guard, mantra-validate) DO NOT fire here. This script therefore:
+//   - NEVER imports/requires _byan/.../lib/fd-state.js and NEVER writes
+//     fd-state.json directly (forbidden by the linter).
+//   - uses NO wall-clock and NO randomness primitive (wall-clock / RNG /
+//     crypto) — they break resume; any timestamp/id arrives via `args`.
+//   - returns DATA only. The orchestrating skill is the human-gated conductor;
+//     IT records FD/strict state via the byan_fd_* / byan_strict_* MCP tools AT
+//     the gate, and IT presents the source step-06 [W]alk/[F]ix/[S]kip menu.
+// The product artifacts (the edited source files, and the tech-spec Status /
+// Review Notes when mode A) are written by the execute/self-check leaves — that
+// is the workflow's output, not BYAN platform state.
+// ---------------------------------------------------------------------------
+
+// Convergence guard — mirror of step-03 prose rule "3 consecutive failures on
+// the same task -> HALT". The sandbox forbids import, so the tiny counter is
+// inlined; it turns the prose into a real cap the model cannot silently overrun.
+const MAX_TASK_CYCLES = 3
+function taskConvergenceGuard({ cycles, passed, maxCycles = MAX_TASK_CYCLES }) {
+  if (passed) return { done: true, halted: false, reason: 'task-green' }
+  if (cycles >= maxCycles) return { done: true, halted: true, reason: `halt: ${maxCycles} consecutive failures on the same task` }
+  return { done: false, halted: false, reason: 'retry' }
+}
+
+// Hard cap on the number of tasks we will drive in one run (the source loops
+// "for each task" with no count; this bounds the outer loop for a resumable run).
+const MAX_TASKS = (args && Number.isInteger(args.maxTasks) && args.maxTasks > 0) ? args.maxTasks : 25
+
+const MODE_SCHEMA = {
+  type: 'object',
+  required: ['executionMode', 'baselineCommit'],
+  properties: {
+    executionMode: { type: 'string', enum: ['tech-spec', 'direct'], description: '"tech-spec" if user gave a spec path, else "direct"' },
+    techSpecPath: { type: 'string', description: 'path to the tech-spec file when executionMode is tech-spec; empty otherwise' },
+    baselineCommit: { type: 'string', description: 'git rev-parse HEAD at start, or "NO_GIT"' },
+    projectContextLoaded: { type: 'boolean', description: 'true if a **/project-context.md was found and loaded' },
+    escalationSignals: { type: 'array', items: { type: 'string' }, description: 'step-01 escalation signals detected in the direct request (multi-component, system-level, uncertainty, multi-layer, extended timeframe)' },
+    escalationLevel: { type: 'integer', description: '0 = none, 0-2 = focused multi-component, 3+ = platform/system work' },
+    escalationAdvice: { type: 'string', enum: ['proceed', 'suggest-plan', 'suggest-bmad'], description: 'what the human gate should consider: proceed = execute directly; suggest-plan = quick-spec; suggest-bmad = full PRD flow' },
+  },
+}
+
+const PLAN_SCHEMA = {
+  type: 'object',
+  required: ['tasks', 'acceptanceCriteria'],
+  properties: {
+    filesToModify: { type: 'array', items: { type: 'string' } },
+    patterns: { type: 'array', items: { type: 'string' }, description: 'code style / conventions / error-handling / test patterns observed' },
+    dependencies: { type: 'array', items: { type: 'string' } },
+    tasks: { type: 'array', items: { type: 'string' }, description: 'ordered task list synthesized from the request' },
+    acceptanceCriteria: { type: 'array', items: { type: 'string' }, description: 'AC inferred from the user request' },
+  },
+}
+
+const TASKS_SCHEMA = {
+  type: 'object',
+  required: ['tasks'],
+  properties: {
+    tasks: { type: 'array', items: { type: 'string' }, description: 'ordered task list to implement' },
+    acceptanceCriteria: { type: 'array', items: { type: 'string' } },
+  },
+}
+
+const VERIFY_SCHEMA = {
+  type: 'object',
+  required: ['passed'],
+  properties: {
+    passed: { type: 'boolean', description: 'true ONLY if this task\'s tests pass with no regressions AND its acceptance criteria are met' },
+    blocking: { type: 'array', items: { type: 'string' }, description: 'blocking issues when not passed' },
+    summary: { type: 'string' },
+  },
+}
+
+const SELF_CHECK_SCHEMA = {
+  type: 'object',
+  required: ['tasksComplete', 'testsPassing', 'acSatisfied', 'patternsFollowed'],
+  properties: {
+    tasksComplete: { type: 'boolean' },
+    testsPassing: { type: 'boolean' },
+    acSatisfied: { type: 'boolean' },
+    patternsFollowed: { type: 'boolean' },
+    filesModified: { type: 'array', items: { type: 'string' } },
+    issues: { type: 'array', items: { type: 'string' }, description: 'any self-audit gaps; empty if clean' },
+    summary: { type: 'string' },
+  },
+}
+
+const FINDING = {
+  type: 'object',
+  required: ['id', 'severity', 'validity', 'description'],
+  properties: {
+    id: { type: 'string', description: 'F1, F2, ...' },
+    severity: { type: 'string', enum: ['Critical', 'High', 'Medium', 'Low'] },
+    validity: { type: 'string', enum: ['real', 'noise', 'undecided'] },
+    description: { type: 'string' },
+  },
+}
+
+const REVIEW_SCHEMA = {
+  type: 'object',
+  required: ['findings'],
+  properties: {
+    findings: { type: 'array', items: FINDING, description: 'ALL findings, ordered by severity (Critical first), numbered F1..Fn; do not drop by severity/validity' },
+    suspiciousZeroFindings: { type: 'boolean', description: 'true if the review returned zero findings (step-05 says this is suspicious and should HALT for human guidance)' },
+    diffSummary: { type: 'string' },
+  },
+}
+
+// --- inputs (everything non-deterministic arrives via args; no clock/RNG) ---
+const userInput = (args && args.request) || (args && args.input) || ''
+const techSpecArg = (args && args.techSpec) || ''
+
+// ===========================================================================
+phase('MODE')
+// step-01: capture baseline, load project-context, classify mode, evaluate the
+// escalation threshold. The human escalation MENU ([P]/[W]/[E]) is NOT executed
+// here — the script records the signal/advice and lets the gate decide.
+const mode = await agent(
+  `You are quick-dev step-01 (mode detection). User input: ${JSON.stringify(userInput)}. ` +
+    `Tech-spec arg (if any): ${JSON.stringify(techSpecArg)}.\n` +
+    `1) Capture baseline: if a git repo, run \`git rev-parse HEAD\` -> baselineCommit; else baselineCommit="NO_GIT". ` +
+    `2) Check for a **/project-context.md and report projectContextLoaded. ` +
+    `3) Classify executionMode: "tech-spec" if a spec path was provided (set techSpecPath), else "direct". ` +
+    `4) For direct mode ONLY, evaluate the escalation signals (multiple components, system-level language, ` +
+    `uncertainty about approach, multi-layer scope, extended timeframe; reduced by simplicity markers / single-file focus). ` +
+    `Use holistic judgment. Set escalationLevel (0 none, 0-2 focused, 3+ platform) and escalationAdvice ` +
+    `(proceed / suggest-plan / suggest-bmad). Do NOT prompt the user — just report.`,
+  { label: 'mode-detection', phase: 'MODE', schema: MODE_SCHEMA }
+)
+log(`mode=${mode.executionMode} baseline=${mode.baselineCommit} escalation=${mode.escalationLevel || 0} advice=${mode.escalationAdvice || 'proceed'}`)
+
+// ===========================================================================
+phase('CONTEXT')
+// step-02: direct mode only — gather files/patterns/deps and synthesize a plan.
+// Mode A (tech-spec) skips this and extracts tasks/AC from the spec instead.
+let plan
+if (mode.executionMode === 'direct') {
+  plan = await agent(
+    `quick-dev step-02 (context gathering, DIRECT mode). Request: ${JSON.stringify(userInput)}.\n` +
+      `Baseline: ${mode.baselineCommit}. project-context loaded: ${Boolean(mode.projectContextLoaded)}.\n` +
+      `Identify the files to modify (glob/grep), the relevant patterns (style, conventions, imports, error handling, test patterns), ` +
+      `the dependencies (external libs, internal modules, config files), then synthesize an ordered task list and the inferred ` +
+      `acceptance criteria. Return the plan; do NOT ask the user y/n/adjust here (that confirmation is the human gate's job).`,
+    { label: 'context-gathering', phase: 'CONTEXT', schema: PLAN_SCHEMA }
+  )
+} else {
+  plan = await agent(
+    `quick-dev (TECH-SPEC mode). Load the tech-spec at ${JSON.stringify(mode.techSpecPath || techSpecArg)} and extract its ` +
+      `tasks/subtasks and acceptance criteria verbatim. Return the ordered task list and the AC.`,
+    { label: 'extract-tech-spec', phase: 'CONTEXT', schema: TASKS_SCHEMA }
+  )
+}
+const tasks = Array.isArray(plan.tasks) ? plan.tasks : []
+log(`tasks=${tasks.length} mode=${mode.executionMode}`)
+
+// ===========================================================================
+phase('EXECUTE')
+// step-03: implement EVERY task without stopping for milestones; per task, run a
+// load->implement->test loop with a 3-failure convergence cap (the prose HALT).
+const taskResults = []
+const runnable = tasks.slice(0, MAX_TASKS)
+for (let i = 0; i < runnable.length; i++) {
+  const task = runnable[i]
+  let cycles = 0
+  let verify = { passed: false, blocking: ['not started'] }
+  let guard = { done: false, halted: false, reason: 'init' }
+  while (true) {
+    cycles += 1
+    const impl = await agent(
+      `quick-dev step-03 execute. Task ${i + 1}/${runnable.length} (attempt ${cycles}): ${JSON.stringify(task)}.\n` +
+        `Mode: ${mode.executionMode}. Plan context: ${JSON.stringify(plan)}.\n` +
+        `Load the files relevant to THIS task, implement following existing patterns and the project-context rules, ` +
+        `handle errors as the codebase does, write tests where appropriate, and run existing tests to catch regressions. ` +
+        `Do NOT mark the task [x] unless its tests actually pass. Report what you changed.`,
+      { label: `execute-task-${i + 1}-cycle-${cycles}`, phase: 'EXECUTE' }
+    )
+    verify = await agent(
+      `Verify quick-dev task ${i + 1}: ${JSON.stringify(task)}. Implementation notes: ${impl}\n` +
+        `Run the relevant tests (infer the command from the repo). passed=true ONLY if this task's tests pass with zero ` +
+        `regressions AND its acceptance criteria are met. Otherwise passed=false with the precise blocking issues.`,
+      { label: `verify-task-${i + 1}-cycle-${cycles}`, phase: 'EXECUTE', schema: VERIFY_SCHEMA }
+    )
+    guard = taskConvergenceGuard({ cycles, passed: Boolean(verify && verify.passed) })
+    log(`task ${i + 1} cycle ${cycles}: passed=${Boolean(verify && verify.passed)} -> ${guard.reason}`)
+    if (guard.done) break
+  }
+  taskResults.push({ index: i + 1, task, passed: Boolean(verify.passed), halted: guard.halted, cycles, blocking: (verify && verify.blocking) || [] })
+  // step-03 HALT condition: stop driving further tasks once a task fails to converge.
+  if (guard.halted) {
+    log(`HALT: task ${i + 1} did not converge after ${MAX_TASK_CYCLES} attempts`)
+    break
+  }
+}
+const allTasksPassed = taskResults.length > 0 && taskResults.every((t) => t.passed)
+const halted = taskResults.some((t) => t.halted)
+
+// ===========================================================================
+phase('SELF_CHECK')
+// step-04: audit tasks/tests/AC/patterns; update tech-spec status when mode A.
+const selfCheck = await agent(
+  `quick-dev step-04 self-check. Mode: ${mode.executionMode}. Tech-spec: ${JSON.stringify(mode.techSpecPath || techSpecArg)}.\n` +
+    `Task results: ${JSON.stringify(taskResults)}. AC: ${JSON.stringify(plan.acceptanceCriteria || [])}.\n` +
+    `Audit: (1) all tasks marked complete with no silent skips; (2) all tests passing, new tests added where needed; ` +
+    `(3) every acceptance criterion demonstrably met incl. edge cases; (4) existing patterns and project-context rules followed. ` +
+    `If mode is "tech-spec": load the spec, mark its tasks [x] and set status "Implementation Complete". Report the audit verdict.`,
+  { label: 'self-check', phase: 'SELF_CHECK', schema: SELF_CHECK_SCHEMA }
+)
+log(`self-check: tasks=${selfCheck.tasksComplete} tests=${selfCheck.testsPassing} ac=${selfCheck.acSatisfied} patterns=${selfCheck.patternsFollowed}`)
+
+// ===========================================================================
+phase('ADVERSARIAL_REVIEW')
+// step-05: build the diff from baselineCommit (incl. new untracked files YOU
+// created), run the adversarial review task, and order + number ALL findings.
+const review = await agent(
+  `quick-dev step-05 adversarial review. Baseline: ${mode.baselineCommit}.\n` +
+    `1) Construct the full diff of changes since the baseline: if a git hash, \`git diff ${mode.baselineCommit}\` PLUS the full ` +
+    `content of any new untracked files YOU created this run (do NOT git add anything; read-only). If "NO_GIT", best-effort diff ` +
+    `of the files you touched plus new files. ` +
+    `2) Run the adversarial review at _byan/command/review-adversarial-general.xml against that diff (inline the task if no ` +
+    `subagent invocation is available). ` +
+    `3) Return ALL findings ordered by severity (Critical first), numbered F1..Fn, each with severity + validity (real/noise/undecided) ` +
+    `+ description. Do NOT drop findings by severity or validity. If the review returns ZERO findings, set suspiciousZeroFindings=true ` +
+    `(step-05 treats zero findings as suspicious and a reason to halt for human guidance).`,
+  { label: 'adversarial-review', phase: 'ADVERSARIAL_REVIEW', schema: REVIEW_SCHEMA }
+)
+const findings = Array.isArray(review.findings) ? review.findings : []
+log(`review: findings=${findings.length} suspiciousZero=${Boolean(review.suspiciousZeroFindings)}`)
+
+// ===========================================================================
+phase('VERDICT')
+// step-06 is a HUMAN decision ([W]alk-through / [F]ix-auto / [S]kip) plus the
+// final tech-spec "Completed" + Review Notes update. We DO NOT pick a resolution
+// here; we hand the gate the ordered, numbered findings and the run summary.
+const realFindings = findings.filter((f) => f && f.validity === 'real')
+const criticalOrHigh = findings.filter((f) => f && (f.severity === 'Critical' || f.severity === 'High'))
+const status = halted
+  ? 'halted-no-convergence'
+  : allTasksPassed && selfCheck.tasksComplete && selfCheck.testsPassing && selfCheck.acSatisfied
+    ? 'review-ready'
+    : 'in-progress'
+
+return {
+  workflow: 'quick-dev',
+  summary: selfCheck.summary || `quick-dev (${mode.executionMode}): ${taskResults.length} task(s) processed, ${findings.length} review finding(s).`,
+  mode: mode.executionMode,
+  baselineCommit: mode.baselineCommit,
+  escalation: { level: mode.escalationLevel || 0, advice: mode.escalationAdvice || 'proceed', signals: mode.escalationSignals || [] },
+  status,
+  steps: 6,
+  tasks: { total: tasks.length, run: taskResults.length, passed: taskResults.filter((t) => t.passed).length, results: taskResults },
+  selfCheck,
+  review: { total: findings.length, real: realFindings.length, criticalOrHigh: criticalOrHigh.length, suspiciousZeroFindings: Boolean(review.suspiciousZeroFindings), findings },
+  // The orchestrating skill presents the step-06 [W]/[F]/[S] resolution menu and,
+  // for mode A, writes the tech-spec "Completed" status + Review Notes at the gate.
+  needsHumanGate: true,
+}

package/install/templates/.claude/workflows/sprint-planning.js

@@ -0,0 +1,261 @@
+export const meta = {
+  name: 'sprint-planning',
+  description: 'Native port of the BYAN sprint-planning workflow: discover and parse all epic files, build the ordered sprint-status structure (epic / stories / retrospective), apply intelligent never-downgrade status detection, write sprint-status.yaml, then validate coverage and report. Returns a structured verdict; the human gate stays in the orchestrating skill.',
+  phases: [
+    { title: 'PARSE', detail: 'discover epic*.md files and extract all epics + stories into kebab-case keys' },
+    { title: 'STRUCTURE', detail: 'build the ordered development_status map: epic, its stories, its retrospective' },
+    { title: 'DETECT', detail: 'intelligent per-story status detection, never downgrade an existing status' },
+    { title: 'GENERATE', detail: 'write sprint-status.yaml with metadata as comments AND as parseable fields' },
+    { title: 'VALIDATE', detail: 'coverage + legal-status + YAML checks, then count totals and report' },
+  ],
+}
+
+// ---------------------------------------------------------------------------
+// FD / STRICT STATE CONTRACT  (re-asserted inline — byan-lint-workflows).
+//
+// The in-CLI Workflow tool runs this script OUTSIDE the conversation turn, so
+// BYAN's main-thread hooks (fd-phase-guard, strict-scope-guard, mantra-
+// validate) DO NOT fire here. This script therefore:
+//   - NEVER imports/requires _byan/.../lib/fd-state.js and NEVER writes
+//     fd-state.json directly. No require/import, no fs, no Date/RNG
+//     (the sandbox forbids them; clock/RNG break resume — any date is passed
+//     in via args.date).
+//   - returns DATA only. The orchestrating skill is the human-gated conductor;
+//     IT records FD/strict state via the byan_fd_* / byan_strict_* MCP tools
+//     AT the gate.
+// The product of THIS workflow is the sprint-status.yaml FILE, written by the
+// GENERATE leaf — that is the artifact, not BYAN platform state.
+// ---------------------------------------------------------------------------
+
+// Inputs are passed in (no clock/RNG inside the sandbox). Mirrors the
+// workflow.yaml variables: epics_location, status_file, project_name, date.
+const epicsLocation = (args && args.epicsLocation) || 'planning artifacts folder (look for *epic*.md or epics/index.md)'
+const statusFile = (args && args.statusFile) || '{implementation_artifacts}/sprint-status.yaml'
+const projectName = (args && args.projectName) || '{project_name from _byan/bmm/config.yaml}'
+const projectKey = (args && args.projectKey) || 'NOKEY'
+const trackingSystem = (args && args.trackingSystem) || 'file-system'
+const storyLocation = (args && args.storyLocation) || '{implementation_artifacts}'
+const date = (args && args.date) || '{date passed by orchestrator}'
+
+// STEP 1 (+ sub-step 0.5 discovery) — Parse epic files and extract all work items.
+// Mirrors instructions.md step n="1" and the FULL_LOAD discovery: whole doc first
+// (epics.md / bmm-epics.md / *epic*.md), else sharded epics/index.md + epic-N.md.
+phase('PARSE')
+const PARSE_SCHEMA = {
+  type: 'object',
+  required: ['epics'],
+  properties: {
+    sourceMode: { type: 'string', enum: ['whole', 'sharded', 'none'], description: 'whole doc, sharded index, or no epics found' },
+    epics: {
+      type: 'array',
+      description: 'every epic discovered, in document order',
+      items: {
+        type: 'object',
+        required: ['num', 'title', 'stories'],
+        properties: {
+          num: { type: 'integer', description: 'epic number from "## Epic N:"' },
+          title: { type: 'string' },
+          stories: {
+            type: 'array',
+            items: {
+              type: 'object',
+              required: ['key', 'title'],
+              properties: {
+                key: { type: 'string', description: 'kebab key: "1-1-user-authentication" (period->dash, title kebab-case)' },
+                title: { type: 'string', description: 'original story title' },
+              },
+            },
+          },
+        },
+      },
+    },
+  },
+}
+const parsed = await agent(
+  `You are the sprint-planning workflow (BYAN, Phase 4 implementation). Source of truth: ` +
+    `_byan/workflow/simple/4-implementation/sprint-planning/instructions.md (steps 0.5 + 1).\n` +
+    `DISCOVER epics under: ${JSON.stringify(epicsLocation)}. FULL_LOAD strategy:\n` +
+    `1. Whole document first — epics.md, bmm-epics.md, user-stories.md, or any *epic*.md (fuzzy match).\n` +
+    `2. Else sharded — read epics/index.md, then read EVERY epic section file it lists (epic-1.md, epic-2.md ...).\n` +
+    `3. If both whole and sharded exist, prefer the whole document.\n` +
+    `For each epic file, EXTRACT:\n` +
+    `- Epic numbers from headers like "## Epic 1:" / "## Epic 2:".\n` +
+    `- Story IDs + titles from "### Story 1.1: User Authentication".\n` +
+    `- Convert each story to a kebab key: replace the period with a dash (1.1 -> 1-1) and kebab-case the title, ` +
+    `e.g. "### Story 1.1: User Authentication" -> "1-1-user-authentication".\n` +
+    `Build the COMPLETE inventory of all epics and stories from ALL epic files. ` +
+    `If NO epic file is found, set sourceMode "none" and return an empty epics array (do not invent epics).`,
+  { label: 'parse-epics', phase: 'PARSE', schema: PARSE_SCHEMA }
+)
+
+// STEP 2 — Build sprint status structure. Mirrors step n="2": per epic emit, in
+// order, epic-{num} (backlog), each story key (backlog), epic-{num}-retrospective
+// (optional). All defaults; status detection happens in DETECT.
+phase('STRUCTURE')
+const STRUCTURE_SCHEMA = {
+  type: 'object',
+  required: ['entries'],
+  properties: {
+    entries: {
+      type: 'array',
+      description: 'ordered development_status entries: epic, its stories, its retrospective, then next epic...',
+      items: {
+        type: 'object',
+        required: ['key', 'kind', 'defaultStatus'],
+        properties: {
+          key: { type: 'string', description: 'epic-N | story kebab key | epic-N-retrospective' },
+          kind: { type: 'string', enum: ['epic', 'story', 'retrospective'] },
+          defaultStatus: { type: 'string', enum: ['backlog', 'optional'], description: 'epic/story=backlog, retrospective=optional' },
+        },
+      },
+    },
+  },
+}
+const structure = await agent(
+  `Step 2 of sprint-planning. From the parsed inventory below, build the ORDERED development_status entry list.\n` +
+    `Inventory: ${JSON.stringify(parsed)}\n` +
+    `For EACH epic, in epic order, emit in this exact order:\n` +
+    `1. epic-{num}  -> kind "epic",  defaultStatus "backlog"\n` +
+    `2. one entry per story, key = the story kebab key -> kind "story", defaultStatus "backlog"\n` +
+    `3. epic-{num}-retrospective -> kind "retrospective", defaultStatus "optional"\n` +
+    `Then move to the next epic. Do NOT add or drop any item relative to the inventory.`,
+  { label: 'build-structure', phase: 'STRUCTURE', schema: STRUCTURE_SCHEMA }
+)
+
+// STEP 3 — Apply intelligent status detection. Mirrors step n="3": for each story
+// check {story_location}/{key}.md -> at least ready-for-dev; if a prior status_file
+// exists with a MORE advanced status, preserve it (NEVER downgrade).
+phase('DETECT')
+const DETECT_SCHEMA = {
+  type: 'object',
+  required: ['entries'],
+  properties: {
+    entries: {
+      type: 'array',
+      items: {
+        type: 'object',
+        required: ['key', 'kind', 'status'],
+        properties: {
+          key: { type: 'string' },
+          kind: { type: 'string', enum: ['epic', 'story', 'retrospective'] },
+          status: { type: 'string', enum: ['backlog', 'ready-for-dev', 'in-progress', 'review', 'done', 'optional'] },
+          reason: { type: 'string', description: 'why this status (file found / preserved from prior status-file / default)' },
+        },
+      },
+    },
+    preservedCount: { type: 'integer', description: 'entries whose status was preserved from a more-advanced prior status-file' },
+  },
+}
+const detected = await agent(
+  `Step 3 of sprint-planning — intelligent status detection. Ordered entries (all defaults):\n` +
+    `${JSON.stringify(structure)}\n` +
+    `Story location: ${JSON.stringify(storyLocation)}. Existing status file (if any): ${JSON.stringify(statusFile)}.\n` +
+    `Rules, applied per entry:\n` +
+    `- STORY: if ${JSON.stringify(storyLocation)}/{key}.md exists, upgrade to at least "ready-for-dev".\n` +
+    `- PRESERVATION (never downgrade): if a prior ${JSON.stringify(statusFile)} exists and holds a MORE advanced status ` +
+    `for this key, KEEP that status. Story order: backlog < ready-for-dev < in-progress < review < done. ` +
+    `Epic order: backlog < in-progress < done. Retrospective: optional <-> done.\n` +
+    `- Items with no signal keep their default (backlog / optional).\n` +
+    `Return every entry with its resolved status and a short reason, preserving the input order. ` +
+    `Count how many statuses were preserved from a prior file in preservedCount.`,
+  { label: 'detect-status', phase: 'DETECT', schema: DETECT_SCHEMA }
+)
+
+// STEP 4 — Generate the sprint status file. Mirrors step n="4" and the template:
+// metadata appears TWICE (as # comments for docs AND as YAML key:value for
+// parsing), then the ordered development_status block.
+phase('GENERATE')
+const GENERATE_SCHEMA = {
+  type: 'object',
+  required: ['written', 'path'],
+  properties: {
+    written: { type: 'boolean', description: 'true once sprint-status.yaml has been written' },
+    path: { type: 'string', description: 'absolute path of the written status file' },
+    metadataDuplicated: { type: 'boolean', description: 'true if metadata appears both as # comments AND as YAML fields' },
+    entryCount: { type: 'integer', description: 'number of development_status lines written' },
+  },
+}
+const generated = await agent(
+  `Step 4 of sprint-planning — write the status file. Resolved entries (preserve order):\n` +
+    `${JSON.stringify(detected)}\n` +
+    `Target: ${JSON.stringify(statusFile)}.\n` +
+    `Write valid YAML following sprint-status-template.yaml. CRITICAL: metadata appears TWICE — ` +
+    `first as a # comment header (generated / project / project_key / tracking_system / story_location, ` +
+    `plus the STATUS DEFINITIONS and WORKFLOW NOTES comment block), then again as parseable YAML key:value fields:\n` +
+    `  generated: ${date}\n  project: ${projectName}\n  project_key: ${projectKey}\n` +
+    `  tracking_system: ${trackingSystem}\n  story_location: ${JSON.stringify(storyLocation)}\n` +
+    `Then a development_status: map with every entry "key: status", in order ` +
+    `(epic, its stories, its retrospective, blank line, next epic...). ` +
+    `Write the complete file to disk and report the path and entry count.`,
+  { label: 'generate-status-file', phase: 'GENERATE', schema: GENERATE_SCHEMA }
+)
+
+// STEP 5 — Validate and report. Mirrors step n="5" + checklist.md: coverage both
+// ways, retrospective per epic, legal statuses, valid YAML, then totals.
+phase('VALIDATE')
+const VALIDATE_SCHEMA = {
+  type: 'object',
+  required: ['passed', 'checks', 'totals'],
+  properties: {
+    passed: { type: 'boolean', description: 'true only if EVERY coverage/legal/YAML check holds' },
+    checks: {
+      type: 'object',
+      required: ['everyEpicPresent', 'everyStoryPresent', 'everyEpicHasRetro', 'noExtraItems', 'allStatusesLegal', 'validYaml'],
+      properties: {
+        everyEpicPresent: { type: 'boolean' },
+        everyStoryPresent: { type: 'boolean' },
+        everyEpicHasRetro: { type: 'boolean' },
+        noExtraItems: { type: 'boolean', description: 'no status-file item absent from the epic files' },
+        allStatusesLegal: { type: 'boolean', description: 'every status value matches the state machine' },
+        validYaml: { type: 'boolean' },
+      },
+    },
+    totals: {
+      type: 'object',
+      required: ['epicCount', 'storyCount'],
+      properties: {
+        epicCount: { type: 'integer' },
+        storyCount: { type: 'integer' },
+        epicsInProgress: { type: 'integer' },
+        storiesDone: { type: 'integer' },
+      },
+    },
+    failures: { type: 'array', items: { type: 'string' }, description: 'specific check failures, empty if passed' },
+  },
+}
+const validation = await agent(
+  `Step 5 of sprint-planning — validate and report. Compare the epic-file inventory against the written status file.\n` +
+    `Inventory: ${JSON.stringify(parsed)}\nResolved entries: ${JSON.stringify(detected)}\nWrite result: ${JSON.stringify(generated)}\n` +
+    `Run every check (per checklist.md):\n` +
+    `- everyEpicPresent: each epic in the epic files appears as epic-N in the status file.\n` +
+    `- everyStoryPresent: each story in the epic files appears (kebab key) in the status file.\n` +
+    `- everyEpicHasRetro: each epic has an epic-N-retrospective entry.\n` +
+    `- noExtraItems: no status-file item is absent from the epic files.\n` +
+    `- allStatusesLegal: every status matches the state machine (epic: backlog/in-progress/done; ` +
+    `story: backlog/ready-for-dev/in-progress/review/done; retro: optional/done).\n` +
+    `- validYaml: the file parses as YAML.\n` +
+    `Count totals: epicCount, storyCount, epicsInProgress, storiesDone. ` +
+    `passed = true only if ALL six checks hold; otherwise list specific failures.`,
+  { label: 'validate-and-report', phase: 'VALIDATE', schema: VALIDATE_SCHEMA }
+)
+
+// Single top-level return — DATA only. The orchestrating skill presents this at
+// the human gate (review the generated file, then track development) and records
+// FD/strict state via MCP. No platform state is touched here.
+return {
+  workflow: 'sprint-planning',
+  summary: `Generated sprint-status from ${parsed.epics.length} epic(s); validation ${validation.passed ? 'PASSED' : 'FAILED'}.`,
+  steps: 5,
+  statusFile: generated.path || statusFile,
+  totals: validation.totals,
+  validationPassed: validation.passed,
+  failures: validation.failures || [],
+  needsHumanGate: true,
+  nextSteps: [
+    'Review the generated sprint-status.yaml',
+    'Use it to track development progress',
+    'Agents update statuses as they work',
+    'Re-run to refresh auto-detected statuses',
+  ],
+  result: { parsed, detected, generated, validation },
+}