@jhizzard/termdeck 0.4.0 → 0.4.2

package/packages/client/public/style.css

@@ -315,10 +315,10 @@
     /* Orchestrator: 1 large left panel (60%), remaining stack on the right (40%). */
     .grid-container.layout-orch {
       grid-template-columns: 3fr 2fr;
-      grid-auto-rows: 1fr;
+      grid-template-rows: repeat(4, 1fr);
     }
     .grid-container.layout-orch .term-panel:first-child {
-      grid-row: 1 / -1;
+      grid-row: 1 / span 4;
       grid-column: 1;
     }
 
@@ -1977,6 +1977,302 @@
       border-color: var(--tg-border-active);
     }
 
+    /* ===== Setup wizard modal (Sprint 19 T2) =====
+       Full-screen overlay with a 4-tier vertical progress rail showing
+       configuration status. Visual language is aligned with .rumen-modal
+       and .add-project-modal so the dashboard feels consistent. */
+    .setup-modal {
+      display: none;
+      position: fixed;
+      inset: 0;
+      z-index: 3200;
+      align-items: center;
+      justify-content: center;
+    }
+    .setup-modal.open { display: flex; }
+    .setup-backdrop {
+      position: absolute;
+      inset: 0;
+      background: rgba(0, 0, 0, 0.72);
+    }
+    .setup-card {
+      position: relative;
+      background: var(--tg-surface);
+      border: 1px solid var(--tg-accent-dim);
+      border-radius: 10px;
+      width: 720px;
+      max-width: calc(100vw - 40px);
+      max-height: calc(100vh - 80px);
+      display: flex;
+      flex-direction: column;
+      box-shadow: 0 16px 48px rgba(0, 0, 0, 0.55);
+      font-family: var(--tg-sans);
+      color: var(--tg-text);
+    }
+    .setup-header {
+      display: flex;
+      align-items: flex-start;
+      justify-content: space-between;
+      gap: 12px;
+      padding: 20px 22px 12px;
+      border-bottom: 1px solid var(--tg-border);
+    }
+    .setup-header h3 {
+      margin: 0 0 4px;
+      font-size: 16px;
+      color: var(--tg-accent);
+    }
+    .setup-subtitle {
+      margin: 0;
+      font-size: 12px;
+      color: var(--tg-text-dim);
+    }
+    .setup-close {
+      background: transparent;
+      border: none;
+      color: var(--tg-text-dim);
+      font-size: 22px;
+      line-height: 1;
+      cursor: pointer;
+      padding: 0 4px;
+      font-family: var(--tg-sans);
+    }
+    .setup-close:hover { color: var(--tg-text-bright); }
+    .setup-body {
+      padding: 16px 22px;
+      overflow-y: auto;
+      flex: 1 1 auto;
+      min-height: 0;
+    }
+    .setup-tiers {
+      display: flex;
+      flex-direction: column;
+    }
+    .setup-loading,
+    .setup-error {
+      padding: 20px 8px;
+      text-align: center;
+      color: var(--tg-text-dim);
+      font-size: 13px;
+    }
+    .setup-error {
+      color: var(--tg-red);
+      text-align: left;
+    }
+    .setup-error code {
+      background: var(--tg-bg);
+      padding: 1px 5px;
+      border-radius: 3px;
+      font-family: var(--tg-mono);
+      font-size: 11px;
+      color: var(--tg-text);
+    }
+    .setup-tier {
+      display: flex;
+      gap: 14px;
+      padding: 4px 0;
+      position: relative;
+    }
+    .setup-tier-rail {
+      display: flex;
+      flex-direction: column;
+      align-items: center;
+      flex-shrink: 0;
+      padding-top: 4px;
+      width: 18px;
+    }
+    .setup-tier-dot {
+      width: 14px;
+      height: 14px;
+      border-radius: 50%;
+      background: var(--tg-border);
+      border: 2px solid var(--tg-bg);
+      box-shadow: 0 0 0 1px var(--tg-border);
+      flex-shrink: 0;
+      z-index: 1;
+    }
+    .setup-tier-line {
+      flex: 1;
+      width: 2px;
+      background: var(--tg-border);
+      margin: 2px 0;
+      min-height: 24px;
+    }
+    .setup-tier-active .setup-tier-dot {
+      background: var(--tg-green);
+      box-shadow: 0 0 0 1px var(--tg-green), 0 0 10px rgba(158, 206, 106, 0.35);
+    }
+    .setup-tier-partial .setup-tier-dot {
+      background: var(--tg-amber);
+      box-shadow: 0 0 0 1px var(--tg-amber);
+    }
+    .setup-tier-not_configured .setup-tier-dot {
+      background: var(--tg-surface-hover);
+      box-shadow: 0 0 0 1px var(--tg-border);
+    }
+    .setup-tier-next .setup-tier-dot {
+      animation: setup-pulse 1.6s ease-in-out infinite;
+    }
+    @keyframes setup-pulse {
+      0%, 100% { box-shadow: 0 0 0 1px var(--tg-accent); }
+      50% { box-shadow: 0 0 0 4px rgba(122, 162, 247, 0.18); }
+    }
+    .setup-tier-body {
+      flex: 1 1 auto;
+      min-width: 0;
+      padding-bottom: 16px;
+    }
+    .setup-tier-head {
+      display: flex;
+      align-items: center;
+      justify-content: space-between;
+      gap: 8px;
+      margin-bottom: 2px;
+    }
+    .setup-tier-name {
+      font-size: 13px;
+      font-weight: 600;
+      color: var(--tg-text-bright);
+    }
+    .setup-tier-status {
+      font-family: var(--tg-mono);
+      font-size: 10px;
+      text-transform: uppercase;
+      letter-spacing: 0.6px;
+      padding: 2px 7px;
+      border-radius: 10px;
+      border: 1px solid var(--tg-border);
+      background: var(--tg-bg);
+      color: var(--tg-text-dim);
+      flex-shrink: 0;
+    }
+    .setup-tier-status-active {
+      color: var(--tg-green);
+      border-color: var(--tg-green);
+      background: rgba(158, 206, 106, 0.08);
+    }
+    .setup-tier-status-partial {
+      color: var(--tg-amber);
+      border-color: var(--tg-amber);
+      background: rgba(224, 175, 104, 0.08);
+    }
+    .setup-tier-status-not_configured {
+      color: var(--tg-text-dim);
+      border-color: var(--tg-border);
+    }
+    .setup-tier-desc {
+      font-size: 12px;
+      color: var(--tg-text-dim);
+      margin-bottom: 4px;
+    }
+    .setup-tier-detail {
+      font-family: var(--tg-mono);
+      font-size: 11px;
+      color: var(--tg-text);
+      margin-bottom: 8px;
+    }
+    .setup-cmds {
+      display: flex;
+      flex-direction: column;
+      gap: 6px;
+      margin-top: 6px;
+    }
+    .setup-cmd {
+      display: flex;
+      align-items: center;
+      gap: 8px;
+      padding: 6px 10px;
+      background: var(--tg-bg);
+      border: 1px solid var(--tg-border);
+      border-radius: var(--tg-radius-sm);
+    }
+    .setup-cmd code {
+      flex: 1 1 auto;
+      min-width: 0;
+      font-family: var(--tg-mono);
+      font-size: 12px;
+      color: var(--tg-cyan);
+      white-space: pre-wrap;
+      word-break: break-word;
+    }
+    .setup-copy {
+      background: transparent;
+      border: 1px solid var(--tg-border);
+      color: var(--tg-text-dim);
+      font-family: var(--tg-sans);
+      font-size: 10px;
+      text-transform: uppercase;
+      letter-spacing: 0.5px;
+      padding: 3px 9px;
+      border-radius: 3px;
+      cursor: pointer;
+      flex-shrink: 0;
+    }
+    .setup-copy:hover {
+      color: var(--tg-text);
+      border-color: var(--tg-border-active);
+    }
+    .setup-copy.copied {
+      color: var(--tg-green);
+      border-color: var(--tg-green);
+    }
+    .setup-footer {
+      display: flex;
+      align-items: center;
+      justify-content: space-between;
+      gap: 12px;
+      padding: 14px 22px 16px;
+      border-top: 1px solid var(--tg-border);
+      flex-wrap: wrap;
+    }
+    .setup-hint {
+      font-size: 11px;
+      color: var(--tg-text-dim);
+      flex: 1 1 auto;
+      min-width: 0;
+    }
+    .setup-hint code {
+      background: var(--tg-bg);
+      padding: 1px 5px;
+      border-radius: 3px;
+      font-family: var(--tg-mono);
+      font-size: 10px;
+      color: var(--tg-text);
+    }
+    .setup-actions {
+      display: flex;
+      gap: 8px;
+      flex-shrink: 0;
+    }
+    .setup-recheck,
+    .setup-done {
+      font-family: var(--tg-sans);
+      font-size: 12px;
+      font-weight: 600;
+      padding: 6px 16px;
+      border-radius: 4px;
+      cursor: pointer;
+      border: 1px solid var(--tg-border);
+    }
+    .setup-recheck {
+      background: transparent;
+      color: var(--tg-text);
+    }
+    .setup-recheck:hover {
+      color: var(--tg-text-bright);
+      border-color: var(--tg-border-active);
+    }
+    .setup-recheck:disabled {
+      opacity: 0.5;
+      cursor: not-allowed;
+    }
+    .setup-done {
+      background: var(--tg-accent);
+      color: var(--tg-bg);
+      border-color: var(--tg-accent);
+    }
+    .setup-done:hover { filter: brightness(1.1); }
+
     /* ===== SCROLLBAR ===== */
     ::-webkit-scrollbar { width: 6px; }
     ::-webkit-scrollbar-track { background: transparent; }

package/packages/server/src/forge-prompt.js

@@ -0,0 +1,265 @@
+// SkillForge — Opus prompt template for crystallizing Mnestra memories into skills.
+//
+// Contract:
+//   systemPrompt           — 4-phase pipeline (audit → extract → generate → self-critique)
+//   buildUserPrompt(mems)  — token-efficient, project-grouped memory bundle
+//   parseSkills(response)  — extract validated skills with evidence + confidence
+
+'use strict';
+
+const SYSTEM_PROMPT = `You are SkillForge, an autonomous knowledge crystallizer. You read a developer's long-term memory (Mnestra) and distill durable, high-signal skills a senior developer would actually reach for. You do NOT summarize memories. You produce skills.
+
+A skill is a reusable playbook: a specific trigger, a concrete procedure, an evidence trail. A memory is raw data. Your job is the conversion.
+
+You run a strict 4-phase pipeline. Skip no phase. Show the decision audit in your reasoning so it can be verified.
+
+=== PHASE 1 — QUALITY AUDIT ===
+
+Score each memory (or cluster of related memories) on actionability from 0.0 to 1.0:
+  - 0.9–1.0  reproducible error→fix pair, precise command sequence, non-obvious gotcha
+  - 0.6–0.8  multi-step procedure with enough detail that a new session could execute it
+  - 0.3–0.5  partial pattern or vague preference; probably needs more evidence
+  - 0.0–0.2  ephemeral task state, one-off observation, sprint meta-process noise
+
+Discard anything below 0.3. Aggressively flag and drop:
+  - Sprint process meta ("T4 follows a gated workflow pattern", "terminal signed off with DONE")
+  - Status snapshots ("current version is v0.4.1") — these decay fast and belong in git/package.json
+  - Generic coding advice ("write tests", "check the config")
+  - Single-occurrence observations with no corroborating evidence
+
+Keep only memories that represent real developer knowledge:
+  - Error → fix pairs (especially ones that burned time)
+  - Multi-step procedures executed 3+ times
+  - Cross-project patterns (a fix in project A applies to project B)
+  - Domain knowledge that is non-obvious to a competent senior dev
+
+=== PHASE 2 — PATTERN EXTRACTION ===
+
+From the surviving memories, identify:
+  - Same error class solved multiple times across projects (high value, high confidence)
+  - Multi-step procedures executed repeatedly (deploy sequences, config rituals, auth dances)
+  - Domain knowledge that is non-obvious (solver formulations, tuning sequences, platform quirks)
+  - Cross-project connections (solution in one repo rescues another)
+
+A pattern must have at least 2 independent supporting memories OR one highly detailed, battle-tested entry. Note which it is.
+
+=== PHASE 3 — SKILL GENERATION ===
+
+For each validated pattern, emit a skill object with fields:
+
+  name           kebab-case, 2–5 words, specific
+  description    one line; explains WHEN this fires, not what it is
+  trigger        specific activating context ("when deploying a Supabase Edge Function", NOT "when coding")
+  body           exact steps, commands, error→fix mappings, gotchas; a senior dev should be able to follow this blind without reopening the source memory
+  evidence       array of memory IDs (or short descriptors) used; the user must be able to audit
+  confidence     0.0–1.0; scales with number of independent sources × detail × recency
+  quality_score  0.0–1.0; your audit score from Phase 1 for the backing cluster
+
+Writing the body:
+  - Lead with the trigger condition in one sentence.
+  - Then a numbered procedure OR an error→fix table.
+  - End with gotchas — the specific thing that bites people.
+  - No fluff. No "it is important to". No closing summary.
+
+Reference skills derived from TermDeck history — these are the calibration bar. Your output should look like these in shape and density.
+
+  --- EXAMPLE SKILL 1: Supabase IPv4 toggle gotcha ---
+  {
+    "name": "supabase-ipv4-connect-toggle",
+    "description": "Fix Supabase Edge Function deploy failing with ECONNREFUSED / direct-connect timeouts on IPv6-only networks",
+    "trigger": "when deploying Supabase Edge Functions, running supabase db push, or hitting ECONNREFUSED against *.supabase.co",
+    "body": "Trigger: supabase CLI commands hang or fail with 'Connection refused' / ENETUNREACH against the direct Postgres host.\\n\\nFix:\\n1. Open Supabase Dashboard → Project → Connect modal (top-right).\\n2. Switch the connection type to 'Session pooler' (IPv4) OR enable the IPv4 add-on for direct connections.\\n3. Copy the pooler connection string (port 5432 → 6543 for transaction pooler, 5432 for session pooler).\\n4. Re-run the deploy / migration.\\n\\nGotcha: the modal's IPv4 toggle is not where it used to be — newer dashboards hide it under 'Connect → Direct connection → IPv4 compatibility'. Most residential ISPs and many CI runners are IPv6-only with broken NAT64, so the default 'Direct connection' silently fails. This bit the Rumen Edge Function deploy during Sprint 3.",
+    "evidence": ["termdeck:rumen-install-guide", "termdeck:sprint-3-deploy-notes", "mnestra:supabase-connect-modal"],
+    "confidence": 0.95,
+    "quality_score": 0.9
+  }
+
+  --- EXAMPLE SKILL 2: Mnestra startup ordering ---
+  {
+    "name": "mnestra-startup-sequence",
+    "description": "Bring Mnestra up cleanly when TermDeck's preflight reports the Mnestra healthz check as red",
+    "trigger": "when TermDeck preflight badge is red for Mnestra, or Mnestra /healthz returns 503, or RAG events are buffering locally without syncing",
+    "body": "Trigger: '[preflight] mnestra unreachable' in server logs, or dashboard health badge shows Mnestra red.\\n\\nSteps (in order — do NOT reorder):\\n1. Verify Postgres (Supabase) is reachable first: psql with the pooler URL from secrets.env. If this fails, stop and fix Postgres before touching Mnestra.\\n2. Confirm pgvector extension exists: SELECT extname FROM pg_extension WHERE extname='vector'. If missing, run the Mnestra migrations via 'termdeck init --mnestra'.\\n3. Start Mnestra server (MCP or HTTP mode per config.yaml rag.mnestra.mode).\\n4. curl http://localhost:<port>/healthz — must return 200 with {ok:true,pg:true,vector:true}.\\n5. Restart TermDeck; preflight re-runs on boot only.\\n\\nGotcha: TermDeck preflight caches its result for the server's lifetime. If you fix Mnestra while TermDeck is running, the badge stays red until a TermDeck restart. Also: the bridge supports three modes (direct/webhook/mcp) — 'direct' needs OPENAI_API_KEY in secrets.env for embedding, the other two do not.",
+    "evidence": ["termdeck:preflight.js", "termdeck:mnestra-bridge", "mnestra:healthz-contract"],
+    "confidence": 0.9,
+    "quality_score": 0.85
+  }
+
+  --- EXAMPLE SKILL 3: Version-drift prevention across the TermDeck/Mnestra/Rumen trio ---
+  {
+    "name": "termdeck-trio-version-bump",
+    "description": "Bump versions across the TermDeck + Mnestra + Rumen trio without creating cross-package drift",
+    "trigger": "when shipping a release that spans @jhizzard/termdeck, @jhizzard/mnestra, or @jhizzard/rumen, or when CLAUDE.md 'Current version' disagrees with package.json",
+    "body": "Trigger: any one of the three packages gets a new version, OR docs reference a version that no longer matches package.json.\\n\\nSteps:\\n1. Decide the semver bump for each package independently — they are NOT lockstep. Only bump a package if its own code changed.\\n2. Update each package.json version field.\\n3. Update every user-facing doc that cites a version: README.md, CLAUDE.md ('Current version' line), docs/GETTING-STARTED.md, docs/INSTALL.md, any launch/* files.\\n4. Grep for the OLD version string across the repo — never trust the 'obvious' list. 'rg -n \"0\\\\.4\\\\.0\"' catches references buried in sprint logs and examples.\\n5. If Mnestra schema changed, bump Mnestra's migration version AND TermDeck's minimum-required-mnestra check in preflight.js.\\n6. Publish in dependency order: mnestra → rumen → termdeck. Consumers break if termdeck ships first referencing an unpublished mnestra.\\n\\nGotcha: CLAUDE.md's 'Current version' line is the #1 source of drift because it's human-maintained and not part of any build step. Also: launch copy (docs/launch/*) freezes the version it was written for — intentionally leave historical launch posts alone, only update evergreen docs.",
+    "evidence": ["termdeck:CLAUDE.md", "termdeck:sprint-19-release", "mnestra:publish-sequence"],
+    "confidence": 0.85,
+    "quality_score": 0.85
+  }
+
+Use these as the density target. If your skills read blander than these, you are not done.
+
+=== PHASE 4 — SELF-CRITIQUE ===
+
+For every generated skill, answer honestly: "Would a senior developer find this genuinely useful, or is this obvious / generic / already in the docs?"
+
+Discard on any YES:
+  - Is this generic coding advice? ("always validate input", "read the config")
+  - Would a senior dev guess this in under 30 seconds without this skill?
+  - Is this a sprint-process observation dressed up as a skill?
+  - Does the body lack a concrete command, path, or error string?
+  - Is the evidence a single vague memory?
+
+Keep only skills where the body contains at least one of: a specific command, a specific file path, a specific error string, or a specific non-obvious configuration detail.
+
+=== OUTPUT FORMAT ===
+
+Emit ONLY a single JSON code block — no prose before or after. Shape:
+
+\`\`\`json
+{
+  "skills": [ /* skill objects, see Phase 3 */ ],
+  "discarded": [
+    { "reason": "sprint process noise", "source": "memory-id-or-descriptor" }
+  ],
+  "notes": "one-paragraph audit summary: how many memories reviewed, how many survived audit, how many skills emitted, overall signal quality"
+}
+\`\`\`
+
+If zero skills pass self-critique, return an empty skills array with notes explaining why. Do NOT invent skills to hit a quota.`;
+
+function buildUserPrompt(memories) {
+  if (!Array.isArray(memories) || memories.length === 0) {
+    return 'No memories supplied. Return {"skills":[],"discarded":[],"notes":"empty input"}.';
+  }
+
+  const byProject = new Map();
+  for (const m of memories) {
+    const project = (m && (m.project || m.source_project)) || 'unscoped';
+    if (!byProject.has(project)) byProject.set(project, []);
+    byProject.get(project).push(m);
+  }
+
+  const lines = [];
+  lines.push(`You are reviewing ${memories.length} memory entries drawn from Mnestra across ${byProject.size} project(s).`);
+  lines.push('Memories are grouped by project. Each entry is: [id] (type, category, recency-days) — content.');
+  lines.push('Run the 4-phase pipeline. Emit the JSON block specified in the system prompt.');
+  lines.push('');
+
+  for (const [project, items] of byProject) {
+    lines.push(`## project: ${project}  (${items.length} memories)`);
+    for (const m of items) {
+      const id = m.id || m.uuid || m.memory_id || '(no-id)';
+      const type = m.type || m.source_type || '-';
+      const category = m.category || '-';
+      const recency = typeof m.age_days === 'number'
+        ? `${m.age_days}d`
+        : (m.created_at ? daysAgo(m.created_at) : '?');
+      const content = oneLine(m.content || m.body || m.text || '');
+      lines.push(`[${id}] (${type}, ${category}, ${recency}) — ${content}`);
+    }
+    lines.push('');
+  }
+
+  lines.push('End of memory bundle. Begin Phase 1 audit.');
+  return lines.join('\n');
+}
+
+function oneLine(s) {
+  return String(s).replace(/\s+/g, ' ').trim().slice(0, 600);
+}
+
+function daysAgo(iso) {
+  const t = Date.parse(iso);
+  if (Number.isNaN(t)) return '?';
+  const d = Math.round((Date.now() - t) / 86400000);
+  return `${d}d`;
+}
+
+function parseSkills(response) {
+  if (!response || typeof response !== 'string') {
+    return { skills: [], discarded: [], notes: 'empty response', raw: response };
+  }
+
+  const block = extractJsonBlock(response);
+  if (!block) {
+    return { skills: [], discarded: [], notes: 'no JSON block found', raw: response };
+  }
+
+  let parsed;
+  try {
+    parsed = JSON.parse(block);
+  } catch (err) {
+    return { skills: [], discarded: [], notes: `JSON parse error: ${err.message}`, raw: block };
+  }
+
+  const skills = Array.isArray(parsed.skills) ? parsed.skills : [];
+  const discarded = Array.isArray(parsed.discarded) ? parsed.discarded : [];
+  const notes = typeof parsed.notes === 'string' ? parsed.notes : '';
+
+  const validated = [];
+  const rejected = [...discarded];
+  for (const s of skills) {
+    const v = validateSkill(s);
+    if (v.ok) {
+      validated.push(v.skill);
+    } else {
+      rejected.push({ reason: v.reason, source: s && s.name });
+    }
+  }
+
+  return { skills: validated, discarded: rejected, notes, raw: block };
+}
+
+function extractJsonBlock(text) {
+  const fenced = text.match(/```json\s*([\s\S]*?)```/i);
+  if (fenced) return fenced[1].trim();
+  const anyFence = text.match(/```\s*([\s\S]*?)```/);
+  if (anyFence && anyFence[1].trim().startsWith('{')) return anyFence[1].trim();
+  const first = text.indexOf('{');
+  const last = text.lastIndexOf('}');
+  if (first !== -1 && last > first) return text.slice(first, last + 1);
+  return null;
+}
+
+function validateSkill(s) {
+  if (!s || typeof s !== 'object') return { ok: false, reason: 'not an object' };
+  const required = ['name', 'description', 'trigger', 'body'];
+  for (const f of required) {
+    if (typeof s[f] !== 'string' || !s[f].trim()) {
+      return { ok: false, reason: `missing field: ${f}` };
+    }
+  }
+  if (!/^[a-z0-9][a-z0-9-]*$/.test(s.name)) {
+    return { ok: false, reason: 'name must be kebab-case' };
+  }
+  const confidence = clamp01(s.confidence);
+  const quality = clamp01(s.quality_score);
+  const evidence = Array.isArray(s.evidence) ? s.evidence.map(String) : [];
+
+  return {
+    ok: true,
+    skill: {
+      name: s.name.trim(),
+      description: s.description.trim(),
+      trigger: s.trigger.trim(),
+      body: s.body,
+      evidence,
+      confidence,
+      quality_score: quality,
+    },
+  };
+}
+
+function clamp01(v) {
+  const n = Number(v);
+  if (!Number.isFinite(n)) return 0;
+  if (n < 0) return 0;
+  if (n > 1) return 1;
+  return n;
+}
+
+module.exports = {
+  systemPrompt: SYSTEM_PROMPT,
+  buildUserPrompt,
+  parseSkills,
+};