kairn-cli 1.10.1 → 1.12.0

package/dist/cli.js

@@ -256,6 +256,113 @@ var ui = {
 `);
   }
 };
+function formatTime(seconds) {
+  if (seconds < 60) return `${seconds}s`;
+  const min = Math.floor(seconds / 60);
+  const sec = seconds % 60;
+  return sec > 0 ? `${min}m ${sec}s` : `${min} min`;
+}
+function estimateTime(model, intent) {
+  const wordCount = intent.split(/\s+/).length;
+  const isComplex = wordCount > 40;
+  const perPass = {
+    "haiku": 5,
+    "sonnet": 20,
+    "opus": 60,
+    "gpt-4.1-mini": 10,
+    "gpt-4.1": 25,
+    "gpt-5": 15,
+    "o4-mini": 12,
+    "gemini-2.5-flash": 8,
+    "gemini-3-flash": 8,
+    "gemini-2.5-pro": 30,
+    "gemini-3.1-pro": 30,
+    "grok-4.1-fast": 10,
+    "grok-4.20": 25,
+    "deepseek": 15,
+    "mistral-large": 20,
+    "codestral": 15,
+    "mistral-small": 10,
+    "llama": 10,
+    "qwen": 10
+  };
+  const basePerPass = Object.entries(perPass).find(([k]) => model.toLowerCase().includes(k))?.[1] ?? 20;
+  const totalBase = basePerPass * 2;
+  if (isComplex) {
+    const low = Math.floor(totalBase * 1.5);
+    const high = Math.floor(totalBase * 4);
+    return `~${formatTime(low)}-${formatTime(high)} (complex workflow)`;
+  }
+  return `~${formatTime(totalBase)}`;
+}
+function createProgressRenderer() {
+  const lines = [];
+  let intervalId = null;
+  let currentPhase = "";
+  let phaseStart = Date.now();
+  let lineCount = 0;
+  function render() {
+    if (lineCount > 0) {
+      process.stdout.write(`\x1B[${lineCount}A`);
+    }
+    for (const line of lines) {
+      process.stdout.write("\x1B[2K" + line + "\n");
+    }
+    lineCount = lines.length;
+  }
+  function updateElapsed() {
+    if (!currentPhase) return;
+    const elapsed = Math.floor((Date.now() - phaseStart) / 1e3);
+    const lastIdx = lines.length - 1;
+    if (lastIdx >= 0) {
+      lines[lastIdx] = lines[lastIdx].replace(/\[\d+s\]/, `[${elapsed}s]`);
+      render();
+    }
+  }
+  return {
+    update(progress) {
+      if (progress.status === "running") {
+        currentPhase = progress.phase;
+        phaseStart = Date.now();
+        lines.push(`  ${warmStone("\u25D0")} ${progress.message} ${chalk.dim("[0s]")}`);
+        if (!intervalId) {
+          intervalId = setInterval(updateElapsed, 1e3);
+        }
+      } else if (progress.status === "success") {
+        const lastIdx = lines.length - 1;
+        const elapsed = progress.elapsed != null ? ` ${chalk.dim("\u2014")} ${chalk.dim(Math.floor(progress.elapsed) + "s")}` : "";
+        const detail = progress.detail ? ` ${chalk.dim("(" + progress.detail + ")")}` : "";
+        if (lastIdx >= 0) {
+          lines[lastIdx] = `  ${chalk.green("\u2714")} ${progress.message}${detail}${elapsed}`;
+        }
+        currentPhase = "";
+      } else if (progress.status === "warning") {
+        const lastIdx = lines.length - 1;
+        if (lastIdx >= 0) {
+          lines[lastIdx] = `  ${chalk.yellow("\u26A0")} ${progress.message}`;
+        }
+        currentPhase = progress.phase;
+        phaseStart = Date.now();
+        lines.push(`  ${warmStone("\u25D0")} Retrying in concise mode... ${chalk.dim("[0s]")}`);
+      }
+      render();
+    },
+    finish() {
+      if (intervalId) clearInterval(intervalId);
+      currentPhase = "";
+      render();
+    },
+    fail(err) {
+      if (intervalId) clearInterval(intervalId);
+      currentPhase = "";
+      const lastIdx = lines.length - 1;
+      if (lastIdx >= 0) {
+        lines[lastIdx] = `  ${chalk.red("\u2716")} Compilation failed`;
+      }
+      render();
+    }
+  };
+}
 
 // src/logo.ts
 import chalk2 from "chalk";
@@ -450,7 +557,6 @@ var initCommand = new Command("init").description("Set up Kairn with your API ke
 import { Command as Command2 } from "commander";
 import { input as input2, confirm, select as select2 } from "@inquirer/prompts";
 import chalk5 from "chalk";
-import ora from "ora";
 
 // src/compiler/compile.ts
 import fs4 from "fs/promises";
@@ -487,7 +593,7 @@ You must output a JSON object matching the SkeletonSpec schema.
 
 - MCP servers: maximum 6. Prefer fewer.
 - Skills: maximum 3. Only include directly relevant ones.
-- Agents: maximum 3. QA pipeline + one specialist.
+- Agents: maximum 5. Orchestration pipeline (/develop) agents.
 - Hooks: maximum 4 (auto-format, block-destructive, PostCompact, plus one contextual).
 
 If the workflow doesn't clearly need a tool, DO NOT include it.
@@ -527,12 +633,12 @@ You must generate all harness content: CLAUDE.md, commands, rules, agents, skill
 ## Core Principles
 
 - **Workflow-specific, not generic.** Every instruction, command, and rule must relate to the user's actual workflow.
-- **Concise CLAUDE.md.** Under 120 lines. No generic text like "be helpful." Include build/test commands, reference docs/ and skills/.
+- **Concise CLAUDE.md.** Under 150 lines. No generic text like "be helpful." Include build/test commands, reference docs/ and skills/.
 - **Security by default.** Always include deny rules for destructive commands and secret file access.
 
 ## CLAUDE.md Template (mandatory structure)
 
-The \`claude_md\` field MUST follow this exact structure (max 120 lines):
+The \`claude_md\` field MUST follow this exact structure (max 150 lines):
 
 \`\`\`
 # {Project Name}
@@ -581,6 +687,25 @@ Use subagents for deep investigation to keep main context clean.
 - Prefer small, focused commits (one feature or fix per commit)
 - Use conventional commits: feat:, fix:, docs:, refactor:, test:
 - Target < 200 lines per PR when possible
+
+## Engineering Standards
+- Lead with answers over reasoning. Be concise.
+- Use absolute file paths in all references.
+- No filler, no inner monologue, no time estimates.
+- Produce load-bearing code \u2014 every line of output should be actionable.
+
+## Tool Usage Policy
+- Prefer Edit tool over sed/awk for file modifications
+- Prefer Grep tool over rg for searching
+- Prefer Read tool over cat for file reading
+- Reserve Bash for: builds, installs, git, network, processes
+- Read and understand existing code before modifying
+- Delete unused code completely \u2014 no compatibility shims
+
+## Code Philosophy
+- Do not create abstractions for one-time operations
+- Complete the task fully \u2014 don't gold-plate, but don't leave it half-done
+- Prefer editing existing files over creating new ones
 \`\`\`
 
 Do not add generic filler. Every line must be specific to the user's workflow.
@@ -589,20 +714,19 @@ Do not add generic filler. Every line must be specific to the user's workflow.
 
 1. A concise, workflow-specific \`claude_md\` (the CLAUDE.md content)
 2. A \`/project:help\` command that explains the environment
-3. A \`/project:tasks\` command for task management via TODO.md
-4. A \`docs/TODO.md\` file for continuity
-5. A \`docs/DECISIONS.md\` file for architectural decisions
-6. A \`docs/LEARNINGS.md\` file for non-obvious discoveries
-7. A \`rules/continuity.md\` rule encouraging updates to DECISIONS.md and LEARNINGS.md
-8. A \`rules/security.md\` rule with essential security instructions
-9. settings.json with deny rules for \`rm -rf\`, \`curl|sh\`, reading \`.env\` and \`secrets/\`
-10. A \`/project:status\` command for code projects (uses ! for live git/test output)
-11. A \`/project:fix\` command for code projects (uses $ARGUMENTS for issue number)
-12. A \`docs/SPRINT.md\` file for sprint contracts (acceptance criteria, verification steps)
-13. A "Verification" section in CLAUDE.md with concrete verify commands for the project
-14. A "Known Gotchas" section in CLAUDE.md (starts empty, grows with corrections)
-15. A "Debugging" section in CLAUDE.md (2 lines: paste raw errors, use subagents)
-16. A "Git Workflow" section in CLAUDE.md (3 rules: small commits, conventional format, <200 lines PR)
+3. A \`docs/DECISIONS.md\` file for architectural decisions
+4. A \`docs/LEARNINGS.md\` file for non-obvious discoveries
+5. A \`rules/continuity.md\` rule encouraging updates to DECISIONS.md and LEARNINGS.md
+6. A \`rules/security.md\` rule with essential security instructions
+7. settings.json with deny rules for \`rm -rf\`, \`curl|sh\`, reading \`.env\` and \`secrets/\`
+8. A \`/project:status\` command for code projects (uses ! for live git/SPRINT.md output)
+9. A \`/project:fix\` command for code projects (uses $ARGUMENTS for issue number)
+10. A \`docs/SPRINT.md\` file as the living spec/plan (replaces TODO.md \u2014 acceptance criteria, verification steps)
+11. A "Verification" section in CLAUDE.md with concrete verify commands for the project
+12. A "Known Gotchas" section in CLAUDE.md (starts empty, grows with corrections)
+13. A "Debugging" section in CLAUDE.md (2 lines: paste raw errors, use subagents)
+14. A "Git Workflow" section in CLAUDE.md (3 rules: small commits, conventional format, <200 lines PR)
+15. "Engineering Standards", "Tool Usage Policy", and "Code Philosophy" sections in CLAUDE.md
 
 ## Shell-Integrated Commands
 
@@ -717,9 +841,10 @@ Merge this into the settings hooks alongside the PreToolUse and PostToolUse hook
 - \`/project:review\` command (review changes)
 - \`/project:test\` command (run and fix tests)
 - \`/project:commit\` command (conventional commits)
-- \`/project:status\` command (live git status, recent commits, TODO overview using ! prefix)
+- \`/project:status\` command (live git status, recent commits, SPRINT.md overview using ! prefix)
 - \`/project:fix\` command (takes $ARGUMENTS as issue number, plans fix, implements, tests, commits)
 - \`/project:sprint\` command (define acceptance criteria before coding, writes to docs/SPRINT.md)
+- \`/project:develop\` command (full development pipeline \u2014 orchestrates @architect \u2192 @planner \u2192 @implementer \u2192 @verifier \u2192 @fixer \u2192 @grill \u2192 @doc-updater through spec, plan, TDD implement, review, and doc update phases)
 - A TDD skill using the 3-phase isolation pattern (RED \u2192 GREEN \u2192 REFACTOR):
   - RED: Write failing test only. Verify it FAILS.
   - GREEN: Write MINIMUM code to pass. Nothing extra.
@@ -729,6 +854,12 @@ Merge this into the settings hooks alongside the PreToolUse and PostToolUse hook
   - \`@qa-orchestrator\` (sonnet) \u2014 delegates to linter and e2e-tester, compiles QA report
   - \`@linter\` (haiku) \u2014 runs formatters, linters, security scanners
   - \`@e2e-tester\` (sonnet, only when Playwright is in tools) \u2014 browser-based QA via Playwright
+- Development pipeline agents (used by /project:develop):
+  - \`@architect\` (opus) \u2014 conducts spec interview with user, writes confirmed spec to docs/SPRINT.md
+  - \`@planner\` (opus) \u2014 reads spec and codebase, creates step-by-step implementation plan in docs/PLAN.md
+  - \`@implementer\` (sonnet) \u2014 TDD-focused implementation, writes failing tests then minimum code to pass
+  - \`@fixer\` (sonnet) \u2014 targeted bug fixing from verifier/review feedback
+  - \`@doc-updater\` (haiku) \u2014 extracts decisions and learnings from completed work, updates docs/DECISIONS.md and docs/LEARNINGS.md
 - \`/project:spec\` command (interview-based spec creation \u2014 asks 5-8 questions one at a time, writes structured spec to docs/SPRINT.md, does NOT start coding until confirmed)
 - \`/project:prove\` command (runs tests, shows git diff vs main, rates confidence HIGH/MEDIUM/LOW with evidence)
 - \`/project:grill\` command (adversarial code review \u2014 challenges each change with "why this approach?", "what if X input?", rates BLOCKER/SHOULD-FIX/NITPICK, blocks until BLOCKERs resolved)
@@ -776,12 +907,12 @@ Return ONLY valid JSON matching this structure:
 
 \`\`\`json
 {
-  "claude_md": "Full CLAUDE.md content (under 120 lines)",
-  "commands": { "help": "...", "tasks": "...", "status": "...", "fix": "...", "sprint": "...", "spec": "...", "prove": "...", "grill": "...", "reset": "..." },
+  "claude_md": "Full CLAUDE.md content (under 150 lines)",
+  "commands": { "help": "...", "develop": "...", "status": "...", "fix": "...", "sprint": "...", "spec": "...", "prove": "...", "grill": "...", "reset": "..." },
   "rules": { "continuity": "...", "security": "..." },
-  "agents": { "qa-orchestrator": "...", "linter": "...", "e2e-tester": "..." },
+  "agents": { "architect": "...", "planner": "...", "implementer": "...", "fixer": "...", "doc-updater": "...", "qa-orchestrator": "...", "linter": "...", "e2e-tester": "..." },
   "skills": { "skill-name/SKILL": "..." },
-  "docs": { "TODO": "...", "DECISIONS": "...", "LEARNINGS": "...", "SPRINT": "..." }
+  "docs": { "DECISIONS": "...", "LEARNINGS": "...", "SPRINT": "..." }
 }
 \`\`\`
 
@@ -798,12 +929,12 @@ You must output a JSON object matching the EnvironmentSpec schema.
 
 - **Minimalism over completeness.** Fewer, well-chosen tools beat many generic ones. Each MCP server costs 500-2000 context tokens.
 - **Workflow-specific, not generic.** Every instruction, command, and rule must relate to the user's actual workflow.
-- **Concise CLAUDE.md.** Under 120 lines. No generic text like "be helpful." Include build/test commands, reference docs/ and skills/.
+- **Concise CLAUDE.md.** Under 150 lines. No generic text like "be helpful." Include build/test commands, reference docs/ and skills/.
 - **Security by default.** Always include deny rules for destructive commands and secret file access.
 
 ## CLAUDE.md Template (mandatory structure)
 
-The \`claude_md\` field MUST follow this exact structure (max 120 lines):
+The \`claude_md\` field MUST follow this exact structure (max 150 lines):
 
 \`\`\`
 # {Project Name}
@@ -852,6 +983,25 @@ Use subagents for deep investigation to keep main context clean.
 - Prefer small, focused commits (one feature or fix per commit)
 - Use conventional commits: feat:, fix:, docs:, refactor:, test:
 - Target < 200 lines per PR when possible
+
+## Engineering Standards
+- Lead with answers over reasoning. Be concise.
+- Use absolute file paths in all references.
+- No filler, no inner monologue, no time estimates.
+- Produce load-bearing code \u2014 every line of output should be actionable.
+
+## Tool Usage Policy
+- Prefer Edit tool over sed/awk for file modifications
+- Prefer Grep tool over rg for searching
+- Prefer Read tool over cat for file reading
+- Reserve Bash for: builds, installs, git, network, processes
+- Read and understand existing code before modifying
+- Delete unused code completely \u2014 no compatibility shims
+
+## Code Philosophy
+- Do not create abstractions for one-time operations
+- Complete the task fully \u2014 don't gold-plate, but don't leave it half-done
+- Prefer editing existing files over creating new ones
 \`\`\`
 
 Do not add generic filler. Every line must be specific to the user's workflow.
@@ -860,20 +1010,19 @@ Do not add generic filler. Every line must be specific to the user's workflow.
 
 1. A concise, workflow-specific \`claude_md\` (the CLAUDE.md content)
 2. A \`/project:help\` command that explains the environment
-3. A \`/project:tasks\` command for task management via TODO.md
-4. A \`docs/TODO.md\` file for continuity
-5. A \`docs/DECISIONS.md\` file for architectural decisions
-6. A \`docs/LEARNINGS.md\` file for non-obvious discoveries
-7. A \`rules/continuity.md\` rule encouraging updates to DECISIONS.md and LEARNINGS.md
-8. A \`rules/security.md\` rule with essential security instructions
-9. settings.json with deny rules for \`rm -rf\`, \`curl|sh\`, reading \`.env\` and \`secrets/\`
-10. A \`/project:status\` command for code projects (uses ! for live git/test output)
-11. A \`/project:fix\` command for code projects (uses $ARGUMENTS for issue number)
-12. A \`docs/SPRINT.md\` file for sprint contracts (acceptance criteria, verification steps)
-13. A "Verification" section in CLAUDE.md with concrete verify commands for the project
-14. A "Known Gotchas" section in CLAUDE.md (starts empty, grows with corrections)
-15. A "Debugging" section in CLAUDE.md (2 lines: paste raw errors, use subagents)
-16. A "Git Workflow" section in CLAUDE.md (3 rules: small commits, conventional format, <200 lines PR)
+3. A \`docs/DECISIONS.md\` file for architectural decisions
+4. A \`docs/LEARNINGS.md\` file for non-obvious discoveries
+5. A \`rules/continuity.md\` rule encouraging updates to DECISIONS.md and LEARNINGS.md
+6. A \`rules/security.md\` rule with essential security instructions
+7. settings.json with deny rules for \`rm -rf\`, \`curl|sh\`, reading \`.env\` and \`secrets/\`
+8. A \`/project:status\` command for code projects (uses ! for live git/SPRINT.md output)
+9. A \`/project:fix\` command for code projects (uses $ARGUMENTS for issue number)
+10. A \`docs/SPRINT.md\` file as the living spec/plan (replaces TODO.md \u2014 acceptance criteria, verification steps)
+11. A "Verification" section in CLAUDE.md with concrete verify commands for the project
+12. A "Known Gotchas" section in CLAUDE.md (starts empty, grows with corrections)
+13. A "Debugging" section in CLAUDE.md (2 lines: paste raw errors, use subagents)
+14. A "Git Workflow" section in CLAUDE.md (3 rules: small commits, conventional format, <200 lines PR)
+15. "Engineering Standards", "Tool Usage Policy", and "Code Philosophy" sections in CLAUDE.md
 
 ## Tool Selection Rules
 
@@ -887,10 +1036,10 @@ Do not add generic filler. Every line must be specific to the user's workflow.
 ## Context Budget (STRICT)
 
 - MCP servers: maximum 6. Prefer fewer.
-- CLAUDE.md: maximum 120 lines.
+- CLAUDE.md: maximum 150 lines.
 - Rules: maximum 5 files, each under 20 lines.
 - Skills: maximum 3. Only include directly relevant ones.
-- Agents: maximum 3. QA pipeline + one specialist.
+- Agents: maximum 5. Orchestration pipeline (/develop) agents.
 - Commands: no limit (loaded on demand, zero context cost).
 - Hooks: maximum 4 (auto-format, block-destructive, PostCompact, plus one contextual).
 
@@ -909,7 +1058,7 @@ Return ONLY valid JSON matching this structure:
     { "tool_id": "id-from-registry", "reason": "why this tool fits" }
   ],
   "harness": {
-    "claude_md": "The full CLAUDE.md content (under 120 lines)",
+    "claude_md": "The full CLAUDE.md content (under 150 lines)",
     "settings": {
       "permissions": {
         "allow": ["Bash(npm run *)", "Read", "Write", "Edit"],
@@ -921,7 +1070,7 @@ Return ONLY valid JSON matching this structure:
     },
     "commands": {
       "help": "markdown content for /project:help",
-      "tasks": "markdown content for /project:tasks"
+      "develop": "markdown content for /project:develop"
     },
     "rules": {
       "continuity": "markdown content for continuity rule",
@@ -931,13 +1080,16 @@ Return ONLY valid JSON matching this structure:
       "skill-name/SKILL": "markdown content with YAML frontmatter"
     },
     "agents": {
-      "qa-orchestrator": "agent markdown with YAML frontmatter"
+      "architect": "agent markdown with YAML frontmatter",
+      "planner": "agent markdown with YAML frontmatter",
+      "implementer": "agent markdown with YAML frontmatter",
+      "fixer": "agent markdown with YAML frontmatter",
+      "doc-updater": "agent markdown with YAML frontmatter"
     },
     "docs": {
-      "TODO": "# TODO\\n\\n- [ ] First task",
       "DECISIONS": "# Decisions\\n\\nArchitectural decisions.",
       "LEARNINGS": "# Learnings\\n\\nNon-obvious discoveries.",
-      "SPRINT": "# Sprint Contract\\n\\nDefine acceptance criteria."
+      "SPRINT": "# Sprint\\n\\nLiving spec and plan."
     }
   }
 }
@@ -1218,7 +1370,7 @@ function buildMcpConfig(skeleton, registry) {
   }
   return config;
 }
-function validateSpec(spec, onProgress) {
+function validateSpec(spec) {
   const warnings = [];
   if (spec.tools.length > 8) {
     warnings.push(`${spec.tools.length} MCP servers selected (recommended: \u22646)`);
@@ -1226,31 +1378,39 @@ function validateSpec(spec, onProgress) {
   if (spec.harness.claude_md) {
     const lines = spec.harness.claude_md.split("\n").length;
     if (lines > 150) {
-      warnings.push(`CLAUDE.md is ${lines} lines (recommended: \u2264100)`);
+      warnings.push(`CLAUDE.md is ${lines} lines (recommended: \u2264150)`);
     }
   }
   if (spec.harness.skills && Object.keys(spec.harness.skills).length > 5) {
     warnings.push(`${Object.keys(spec.harness.skills).length} skills (recommended: \u22643)`);
   }
-  for (const warning of warnings) {
-    onProgress?.(`\u26A0 ${warning}`);
-  }
+  return warnings;
 }
 async function compile(intent, onProgress) {
+  const startTime = Date.now();
   const config = await loadConfig();
   if (!config) {
     throw new Error("No config found. Run `kairn init` first.");
   }
-  onProgress?.("Loading tool registry...");
+  onProgress?.({ phase: "registry", status: "running", message: "Loading tool registry..." });
   const registry = await loadRegistry();
-  onProgress?.("Analyzing workflow...");
+  onProgress?.({ phase: "registry", status: "success", message: "Tool registry loaded", detail: `${registry.length} tools` });
+  onProgress?.({ phase: "pass1", status: "running", message: "Pass 1: Analyzing workflow & selecting tools..." });
   const skeletonMsg = buildSkeletonMessage(intent, registry);
   const skeletonText = await callLLM(config, skeletonMsg, {
     maxTokens: 2048,
     systemPrompt: SKELETON_PROMPT
   });
   const skeleton = parseSkeletonResponse(skeletonText);
-  onProgress?.("Generating environment...");
+  const toolNames = skeleton.tools.map((t) => t.tool_id).join(", ");
+  onProgress?.({
+    phase: "pass1",
+    status: "success",
+    message: `Pass 1: Selected ${skeleton.tools.length} tools`,
+    detail: toolNames,
+    elapsed: (Date.now() - startTime) / 1e3
+  });
+  onProgress?.({ phase: "pass2", status: "running", message: "Pass 2: Generating CLAUDE.md, commands, agents..." });
   const harnessMsg = buildHarnessMessage(intent, skeleton);
   let harness;
   try {
@@ -1260,7 +1420,7 @@ async function compile(intent, onProgress) {
     });
     harness = parseHarnessResponse(harnessText);
   } catch {
-    onProgress?.("Retrying with concise mode...");
+    onProgress?.({ phase: "pass2-retry", status: "warning", message: "Pass 2: Response too large, retrying in concise mode..." });
     const retryMsg = buildHarnessMessage(intent, skeleton, true);
     const retryText = await callLLM(config, retryMsg, {
       maxTokens: 8192,
@@ -1268,9 +1428,19 @@ async function compile(intent, onProgress) {
     });
     harness = parseHarnessResponse(retryText);
   }
-  onProgress?.("Configuring tools...");
+  const cmdCount = Object.keys(harness.commands).length;
+  const agentCount = Object.keys(harness.agents ?? {}).length;
+  const ruleCount = Object.keys(harness.rules).length;
+  onProgress?.({
+    phase: "pass2",
+    status: "success",
+    message: `Pass 2: Generated ${cmdCount} commands, ${agentCount} agents, ${ruleCount} rules`,
+    elapsed: (Date.now() - startTime) / 1e3
+  });
+  onProgress?.({ phase: "pass3", status: "running", message: "Pass 3: Configuring MCP servers & settings..." });
   const settings = buildSettings(skeleton, registry);
   const mcpConfig = buildMcpConfig(skeleton, registry);
+  onProgress?.({ phase: "pass3", status: "success", message: "Pass 3: Configured MCP servers & settings" });
   const spec = {
     id: `env_${crypto.randomUUID()}`,
     intent,
@@ -1290,7 +1460,12 @@ async function compile(intent, onProgress) {
       docs: harness.docs
     }
   };
-  validateSpec(spec, onProgress);
+  const warnings = validateSpec(spec);
+  for (const w of warnings) {
+    onProgress?.({ phase: "done", status: "warning", message: `\u26A0 ${w}` });
+  }
+  const totalElapsed = ((Date.now() - startTime) / 1e3).toFixed(0);
+  onProgress?.({ phase: "done", status: "success", message: `Environment compiled in ${totalElapsed}s`, elapsed: (Date.now() - startTime) / 1e3 });
   await ensureDirs();
   const envPath = path4.join(getEnvsDir(), `${spec.id}.json`);
   await fs4.writeFile(envPath, JSON.stringify(spec, null, 2), "utf-8");
@@ -1345,10 +1520,10 @@ Read .claude/commands/ and list each one with a one-line description.
 Group them by workflow phase:
 
   PLAN:    /project:spec, /project:sprint, /project:plan
-  BUILD:   (just start coding \u2014 Claude reads CLAUDE.md automatically)
+  BUILD:   /project:develop (full pipeline), or just start coding
   VERIFY:  /project:prove, /project:grill, /project:test
   SHIP:    /project:commit, /project:review
-  MANAGE:  /project:status, /project:tasks, /project:reset
+  MANAGE:  /project:status, /project:reset
 
 ## Your Agents
 Read .claude/agents/ and explain each one with how to invoke it.
@@ -1407,7 +1582,7 @@ var LOOP_COMMAND_CODE = `# Development Loop
 Run an assisted development cycle for the next feature.
 
 ## Phase 1: SPEC
-Review docs/TODO.md and docs/SPRINT.md.
+Review docs/SPRINT.md.
 If no sprint is defined, run /project:spec to interview the user.
 Wait for user approval of the spec.
 
@@ -1432,7 +1607,7 @@ Fix any BLOCKERs.
 
 ## Phase 6: SHIP
 Run /project:commit.
-Report what was built and what's next from docs/TODO.md.
+Report what was built and what's next from docs/SPRINT.md.
 
 Then ask: "Continue to next feature?"
 If yes, return to Phase 1.`;
@@ -1441,7 +1616,7 @@ var LOOP_COMMAND_RESEARCH = `# Research Loop
 Run an assisted research cycle.
 
 ## Phase 1: QUESTION
-Review docs/TODO.md for the next research question.
+Review docs/SPRINT.md for the next research question.
 If none, ask the user what to investigate.
 
 ## Phase 2: RESEARCH
@@ -1457,7 +1632,7 @@ Present the summary. Ask the user for feedback.
 Revise based on feedback.
 
 ## Phase 5: NEXT
-Update docs/TODO.md \u2014 mark question as done, identify follow-ups.
+Update docs/SPRINT.md \u2014 mark question as done, identify follow-ups.
 Ask: "Continue to next question?"`;
 var PM_AGENT = `---
 name: pm
@@ -1468,7 +1643,7 @@ model: opus
 You are a project manager for this codebase.
 
 Your responsibilities:
-1. Maintain docs/TODO.md \u2014 keep it prioritized and current
+1. Maintain docs/SPRINT.md \u2014 keep it prioritized and current
 2. Write specs to docs/SPRINT.md when asked
 3. Review completed work and suggest what's next
 4. Track decisions in docs/DECISIONS.md
@@ -1487,7 +1662,7 @@ PM-driven development loop with PR delivery.
 
 ## Phase 1: PLAN (@pm)
 Use @pm to:
-- Read docs/TODO.md and docs/SPRINT.md
+- Read docs/SPRINT.md
 - Select the highest-priority unfinished task
 - Write a spec to docs/SPRINT.md
 - Present the spec for approval
@@ -1517,7 +1692,7 @@ Create a pull request:
 ## Phase 6: NEXT
 Report:
   "PR #{N} ready for review: {link}
-   Next priority from TODO.md: {next task}
+   Next priority from SPRINT.md: {next task}
    Continue? (y/n)"
 
 If yes, return to Phase 1 with next task.`;
@@ -1534,7 +1709,7 @@ PRs are opened automatically. You review when ready.
 
 ## The Loop
 Repeat until max features reached or stopped:
-1. @pm selects next priority from docs/TODO.md
+1. @pm selects next priority from docs/SPRINT.md
 2. Create worktree + branch
 3. Implement the feature
 4. Run verification (build, test, lint)
@@ -1606,7 +1781,7 @@ function autonomyLabel(level) {
 
 // src/adapter/claude-code.ts
 var STATUS_LINE = {
-  command: `printf '%s | %s tasks' "$(git branch --show-current 2>/dev/null || echo 'no-git')" "$(grep -c '\\- \\[ \\]' docs/TODO.md 2>/dev/null || echo 0)"`
+  command: `printf '%s | %s tasks' "$(git branch --show-current 2>/dev/null || echo 'no-git')" "$(grep -c '\\- \\[ \\]' docs/SPRINT.md 2>/dev/null || echo 0)"`
 };
 function isCodeProject(spec) {
   const commands = spec.harness.commands ?? {};
@@ -2066,16 +2241,19 @@ ${clarificationLines}`;
 Autonomy level: ${autonomyLevel} (${autonomyLabel(autonomyLevel)})`;
   }
   console.log(ui.section("Compilation"));
-  const spinner = ora({ text: "Loading tool registry...", indent: 2 }).start();
+  const estimate = estimateTime(config.model, finalIntent);
+  console.log(chalk5.dim(`  Estimated time: ${estimate} (${config.model})`));
+  console.log("");
+  const renderer = createProgressRenderer();
   let spec;
   try {
-    spec = await compile(finalIntent, (msg) => {
-      spinner.text = msg;
+    spec = await compile(finalIntent, (progress) => {
+      renderer.update(progress);
     });
     spec.autonomy_level = autonomyLevel;
-    spinner.succeed("Environment compiled");
+    renderer.finish();
   } catch (err) {
-    spinner.fail("Compilation failed");
+    renderer.fail(err);
     const msg = err instanceof Error ? err.message : String(err);
     console.log(chalk5.red(`
   ${msg}
@@ -2320,7 +2498,7 @@ var updateRegistryCommand = new Command5("update-registry").description("Fetch t
 import { Command as Command6 } from "commander";
 import { confirm as confirm2 } from "@inquirer/prompts";
 import chalk9 from "chalk";
-import ora2 from "ora";
+import ora from "ora";
 import fs12 from "fs/promises";
 import path12 from "path";
 
@@ -2645,7 +2823,7 @@ var optimizeCommand = new Command6("optimize").description("Scan an existing pro
   }
   const targetDir = process.cwd();
   console.log(ui.section("Project Scan"));
-  const scanSpinner = ora2({ text: "Scanning project...", indent: 2 }).start();
+  const scanSpinner = ora({ text: "Scanning project...", indent: 2 }).start();
   const profile = await scanProject(targetDir);
   scanSpinner.stop();
   if (profile.language) console.log(ui.kv("Language:", profile.language));
@@ -2714,10 +2892,10 @@ var optimizeCommand = new Command6("optimize").description("Scan an existing pro
   }
   const intent = buildOptimizeIntent(profile);
   let spec;
-  const spinner = ora2({ text: "Compiling optimized environment...", indent: 2 }).start();
+  const spinner = ora({ text: "Compiling optimized environment...", indent: 2 }).start();
   try {
-    spec = await compile(intent, (msg) => {
-      spinner.text = msg;
+    spec = await compile(intent, (progress) => {
+      spinner.text = progress.message;
     });
     spinner.succeed("Environment compiled");
   } catch (err) {