bigpowers 2.87.1 → 2.87.3

package/VALIDACAO_CONTRA_GSD.md

@@ -0,0 +1,321 @@
+# Validação Cruzada: Bigpowers vs. GSD (Get Shit Done)
+
+**Data:** 2026-05-18  
+**Scope:** Comparar consolidação proposta com arquitetura de artefatos de GSD
+
+---
+
+## 1. Comparação de Estrutura
+
+### GSD (Get Shit Done)
+```
+.planning/
+├── PROJECT.md       ← Contexto do projeto (validado vs. em escopo)
+├── ROADMAP.md       ← Fases + plans + success criteria
+├── STATE.md         ← Estado atual (< 100 linhas)
+├── {phase}-{plan}-PLAN.md    ← Planos individuais
+└── todos/pending/            ← Ideias capturadas
+
+templates/             ← 30+ templates
+references/            ← 60+ docs de referência técnica
+workflows/             ← 93+ workflows (comandos CLI)
+```
+
+### Bigpowers (Proposta de Consolidação)
+```
+specs/
+├── STATE.md              ← Estado atual (59 linhas) ✅
+├── RELEASE-PLAN.md       ← Roadmap vivo ✅
+├── PLAN-v*.md            ← Histórico → archive/ ✅
+├── archive/releases/     ← Novo: histórico consolidado
+├── CONTEXT.md, UBIQUITOUS_LANGUAGE.md, etc.
+└── ...
+
+SKILL-INDEX.md           ← Novo: tabela canônica
+README.md                ← Simplificado
+CONVENTIONS.md           ← Existente
+```
+
+---
+
+## 2. Gap Analysis: O Que Falta em Bigpowers
+
+### 2.1 Falta: PROJECT.md (Contexto do Projeto)
+
+**GSD tem:**
+```markdown
+# PROJECT.md
+
+## What This Is
+[Descrição atual do projeto]
+
+## Core Value
+[Uma coisa que importa mais]
+
+## Requirements
+- Validated (já enviado, comprovado)
+- Active (em construção)
+- Out of Scope (com razão do exclusão)
+
+## Context
+[Background técnico e de usuário]
+
+## Constraints
+[Limitações: stack, timeline, etc.]
+
+## Key Decisions
+[Histórico de decisões significativas]
+```
+
+**Bigpowers não tem equivalente.** Tem CLAUDE.md (comandos, stack, arquitetura), mas não tem:
+- ❌ Requisitos (Validated/Active/Out of Scope)
+- ❌ Core Value (a uma coisa que importa)
+- ❌ Decisões significativas rastreadas
+- ❌ Constraints formalizadas
+
+**Recomendação:** Criar `specs/PROJECT.md` para bigpowers com:
+- What This Is: "38 agent skills for spec-driven TDD by solo developers"
+- Core Value: "Maintain quality (audit score) while scaling skill count"
+- Active Requirements: Feature completeness, testing mandates, etc.
+- Out of Scope: Domain-specific skills (scaffold-exercises), plugins, marketplace
+- Key Decisions: verb-noun naming, local-first (sem GitHub issues), BMAD phases, etc.
+
+---
+
+### 2.2 ROADMAP.md vs. RELEASE-PLAN.md
+
+**GSD usa ROADMAP.md:**
+- Fases numeradas (1, 2, 3, ...)
+- Decimal phases para inserts (2.1 = "critical fix inserted between 2 and 3")
+- Success Criteria (observable behaviors)
+- Plans count per phase
+- Progress table com milestone column
+
+**Bigpowers usa RELEASE-PLAN.md:**
+- Releases (v1.11, v1.12, ...)
+- WSJF scoring
+- Status (✅, ⏳)
+- Focus area (Benchmarks, Auditor, Karpathy, etc.)
+- Detailed action items per release
+
+**Análise:**
+- ✅ RELEASE-PLAN.md é mais rico que ROADMAP (tem WSJF, status, detalhes)
+- ✅ Ambas rastreiam progresso
+- ⚠️ Bigpowers não tem "Success Criteria" formalizados (deveriam estar em STATE ou PROJECT)
+
+**Recomendação:** Manter RELEASE-PLAN.md, mas adicionar "Success Criteria" por release:
+```markdown
+### v1.17.0 — Guardrails (WSJF 3.2) ⏳
+
+**Success Criteria:**
+- [ ] "Zoom-out mandate" implemented in plan-work as HARD-GATE
+- [ ] Surgical changes checklist in audit-code passes 100% of tests
+- [ ] CONVENTIONS.md explicitly forbids "touch unrelated code"
+```
+
+---
+
+### 2.3 Artefatos Intermediários (Templates)
+
+**GSD tem 30+ templates:**
+- AI-SPEC.md, DEBUG.md, VALIDATION.md
+- discovery.md, research.md, spec.md
+- planner-subagent-prompt.md
+- continue-here.md
+
+**Bigpowers tem:**
+- Implícitos em cada skill (não há templates separados)
+- Rely em CONVENTIONS.md
+
+**Recomendação:** Criar `templates/` subdir em bigpowers:
+```
+templates/
+├── spec.md              (→ specs/PLAN.md output format)
+├── phase-discussion.md  (→ output de elaborate-spec)
+├── diagnosis.md         (→ output de investigate-bug)
+└── ...
+```
+
+---
+
+### 2.4 Referências Técnicas (References)
+
+**GSD tem 60+ references:**
+- gates.md (quais gates bloquear, quais permitir)
+- planner-antipatterns.md (o que não fazer)
+- verification-patterns.md (como testar)
+- thinking-models-*.md (modelos de pensamento por agente)
+- tdd.md, executor-examples.md
+
+**Bigpowers tem:**
+- AUDIT.md (auditoria de cobertura de Clean Code)
+- COMPARISON.md (comparação com superpowers, BMAD, etc.)
+- Skills source em SKILL.md (inline docs)
+
+**Recomendação:** Segregar em `references/`:
+```
+references/
+├── clean-code-mapping.md     (como cada skill implementa Clean Code)
+├── gates-and-guardrails.md   (HARD-GATEs em cada skill)
+├── skill-antipatterns.md     (o que não fazer em cada skill)
+├── bmad-lifecycle.md         (phases BMAD detalhadas)
+└── ...
+```
+
+---
+
+## 3. Validação: Minha Proposta vs. GSD
+
+### O Que Está Certo em Minha Proposta
+
+✅ **Hierarquia de artefatos**
+- STATE → RELEASE-PLAN → SKILL-INDEX → README
+- GSD faz: PROJECT → ROADMAP → STATE → PLAN
+- Abordagens equivalentes, estilos diferentes
+
+✅ **Consolidação de redundância**
+- Deletar PLAN.md (stale) ← mesma que GSD faria
+- Arquivar PLAN-v*.md ← GSD collapsa em `<details>` tags
+- Single skill index ← GSD não tem equivalent, é um win para bigpowers
+
+✅ **Nomes simples (verb-noun)**
+- Bigpowers: `develop-tdd`, `validate-fix`, etc.
+- GSD: `/gsd-new-project`, `/gsd-plan-phase`, etc. (verb-noun style)
+- ✓ Bigpowers está bem aqui
+
+✅ **Clareza de fases (BMAD)**
+- Bigpowers: Discover → Elaborate → Plan → Build → Sustain
+- GSD: Phase 1, 2, 3, ... (fases numeradas)
+- Ambas funcionam; BMAD é mais descritivo
+
+### O Que Está Incompleto em Minha Proposta
+
+❌ **Falta PROJECT.md**
+- GSD começa aqui: "What This Is", "Core Value", "Requirements"
+- Bigpowers tem CLAUDE.md, mas não `specs/PROJECT.md`
+- **Ação:** Adicionar PROJECT.md como canônico de contexto do projeto
+
+❌ **Falta Success Criteria formalizados**
+- GSD em ROADMAP: "Observable behaviors that must be TRUE"
+- Bigpowers em RELEASE-PLAN: Apenas "Focus Area" e "Objective"
+- **Ação:** Adicionar Success Criteria por release em RELEASE-PLAN.md
+
+❌ **Falta templates/ e references/**
+- GSD segrega templates (30+) e references (60+)
+- Bigpowers tem tudo inline em SKILL.md
+- **Ação:** Opcional; se fizer, criar `templates/` e `references/` como em GSD
+
+❌ **Falta "Skill Antipatterns" ou "Gates"**
+- GSD tem gates.md (quais gates bloquear)
+- Bigpowers tem mandates em prose, não formalizados
+- **Ação:** Criar `references/gates-and-guardrails.md`
+
+---
+
+## 4. Recomendação: Incorporar Melhores Práticas de GSD
+
+### Fase 1: Imediato (Incorporar em Análise/SKILL-INDEX)
+
+- [ ] Criar `specs/PROJECT.md` com:
+  - What This Is
+  - Core Value
+  - Active Requirements
+  - Out of Scope
+  - Key Decisions (a tabela de ADRs existentes)
+
+- [ ] Atualizar RELEASE-PLAN.md com Success Criteria por release
+
+- [ ] Manter SKILL-INDEX.md como proposto (similar a nada em GSD, é um win único)
+
+### Fase 2: Médio Prazo (30 min de trabalho)
+
+- [ ] Criar `templates/` com estruturas esperadas (spec.md, diagnosis.md, etc.)
+
+- [ ] Criar `references/bmad-phases.md` (ciclo BMAD detalhado)
+
+- [ ] Criar `references/gates-and-guardrails.md` (HARD-GATEs por skill)
+
+### Fase 3: Longo Prazo (Nice-to-Have)
+
+- [ ] Criar `references/skill-antipatterns.md` (o que não fazer)
+
+- [ ] Criar `references/clean-code-mapping.md` (como cada skill implementa Clean Code)
+
+---
+
+## 5. Tabela: Estrutura Final Recomendada
+
+| Nível | Artefato | Propósito | GSD Equivalente |
+|---|---|---|---|
+| **Visão** | specs/PROJECT.md | Contexto, requisitos, decisões | PROJECT.md |
+| **Roadmap** | specs/RELEASE-PLAN.md | Sequência de releases + success criteria | ROADMAP.md |
+| **Estado** | specs/STATE.md | Posição atual (< 100 linhas) | STATE.md |
+| **Skill Catalog** | SKILL-INDEX.md | Tabela canônica de 38 skills | (N/A em GSD) |
+| **Templates** | templates/*.md | Estruturas esperadas | templates/ |
+| **Referência** | references/*.md | Gates, antipatterns, ciclo BMAD | references/ |
+| **Documentação** | README.md | Install, quickstart | README.md |
+| **Convênios** | CONVENTIONS.md | Code, test, git, specs/ standards | CONVENTIONS.md |
+| **Histórico** | specs/archive/releases/PLAN-v*.md | Planos completados | (Collapsed em ROADMAP) |
+
+---
+
+## 6. Validação Final: Aplicar ao Projeto Skills
+
+**Antes (Atual):**
+```
+PLAN.md (stale, 453 linhas)
+        ↓
+specs/RELEASE-PLAN.md (vivo, 104 linhas)
+      ↓
+specs/STATE.md (59 linhas) ✓
+      ↓
+specs/PLAN-v*.md (16 arquivos, dispersos)
+      ↓
+README.md (191 linhas, copia RELEASE-PLAN)
+```
+
+**Depois (Proposto + GSD Best Practice):**
+```
+specs/PROJECT.md (novo)           ← Core Value, Requirements, Decisions
+        ↓
+specs/RELEASE-PLAN.md (revisado)  ← Roadmap + Success Criteria
+        ↓
+specs/STATE.md (mantém)           ← Current Position
+        ↓
+SKILL-INDEX.md (novo)             ← Tabela canônica
+        ↓
+README.md (simplificado)          ← Install, quickstart, link a SKILL-INDEX
+        ↓
+templates/ (novo, opcional)       ← Estruturas esperadas
+references/ (novo, opcional)      ← Gates, antipatterns, BMAD phases
+```
+
+**Resultado:**
+- ✅ Sem redundância entre PLAN.md, RELEASE-PLAN, README
+- ✅ Hierarquia clara: PROJECT → RELEASE-PLAN → STATE → SKILL-INDEX
+- ✅ Incorpora melhores práticas de GSD sem copiar excessivamente
+- ✅ Nomes simples (verb-noun) mantidos
+- ✅ Fases BMAD explícitas em PROJECT.md ou documento separado
+
+---
+
+## 7. Checklist Final
+
+- [x] Minha análise identificou redundâncias corretas (PLAN.md, PLAN-v*.md, skill index)
+- [x] Minha hierarquia proposta é consistente com GSD (STATE → ROADMAP → skills)
+- [x] Faltou: PROJECT.md (contexto do projeto)
+- [x] Faltou: Success Criteria formalizados
+- [x] Faltou: templates/ e references/ (nice-to-have)
+- [x] SKILL-INDEX.md é único valor adicionado vs. GSD (bom!)
+- [x] Nomes simples e ciclo BMAD estão corretos
+- [x] Recomendação: Incorporar PROJECT.md + Success Criteria antes de executar
+
+---
+
+## 8. Próximos Passos Imediatos
+
+1. ✅ Criar `specs/PROJECT.md` com estrutura de GSD
+2. ✅ Atualizar `specs/RELEASE-PLAN.md` com Success Criteria
+3. ✅ Finalmente: Deletar `PLAN.md`, arquivar `PLAN-v*.md`, simplificar README
+4. 📝 Opcional: Criar `templates/` e `references/` conforme GSD
+

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bigpowers",
-  "version": "2.87.1",
+  "version": "2.87.3",
   "description": "73 agent skills synthesizing 17 years of software engineering discipline into a prescriptive methodology for solo developers",
   "main": "index.js",
   "scripts": {

package/scripts/lib/git-hours.sh

@@ -18,7 +18,7 @@ partition() {
   git -C "$ROOT" log --no-merges "$range" \
       --pretty=format:"%at%x1f%H%x1f%(trailers:key=${key},valueonly,separator=%x2C)" \
     | sort -n \
-    | awk -F '\x1f' -v IDLE="$IDLE_SECONDS" -v PAD="$PAD_SECONDS" '
+    | awk -v FS="$(printf '\037')" -v IDLE="$IDLE_SECONDS" -v PAD="$PAD_SECONDS" '
         {
           ts = $1 + 0; story = $3
           gsub(/[[:space:]]/, "", story)

package/scripts/lib/skill-common.sh

@@ -48,6 +48,9 @@ parse_frontmatter() {
     fi
     [ "$in_block" -ne 1 ] && continue
 
+    # Skip comment lines (e.g. # story: tags inside the frontmatter block)
+    [[ "$line" == \#* || "$line" == \<\!--* ]] && continue
+
     case "$line" in
       name:*)
         name="${line#name: }"
@@ -76,6 +79,8 @@ parse_frontmatter() {
       continue
     fi
     [ "$continued" -eq 0 ] && continue
+    # Skip comment lines (e.g. # story: tags inside the frontmatter block)
+    [[ "$line" == \#* || "$line" == \<\!--* ]] && continue
     # If next field starts (unindented key:), stop
     [[ "$line" =~ ^[a-zA-Z_]+: ]] && break
     desc="$desc $line"
@@ -103,7 +108,7 @@ parse_frontmatter_okf() {
   _PF_NAME=$(awk '/^---/{f++} f==1 && /^name:/{print; exit}' "$file" | sed 's/^name:[[:space:]]*//')
   _PF_MODEL=$(awk '/^---/{f++} f==1 && /^model:/{print; exit}' "$file" | sed 's/^model:[[:space:]]*//')
   _PF_EFFORT=$(awk '/^---/{f++} f==1 && /^effort:/{print; exit}' "$file" | sed 's/^effort:[[:space:]]*//')
-  _PF_DESCRIPTION=$(awk '/^---/{f++; next} f==1 && /^description:/{p=1; sub(/^description:[[:space:]]*/,""); print; next} f==1 && p && /^[a-z]+:/{exit} f==1 && p{print}' "$file" \
+  _PF_DESCRIPTION=$(awk '/^---/{f++; next} f==1 && /^description:/{p=1; sub(/^description:[[:space:]]*/,""); print; next} f==1 && p && /^[a-z]+:/{exit} f==1 && p && !/^[[:space:]]*#/ && !/^[[:space:]]*<!--/{print}' "$file" \
     | tr -d '\n' \
     | sed -E 's/[[:space:]]+/ /g')
 

package/scripts/record-cycle-time.sh

@@ -90,7 +90,7 @@ cmd_report() {
 }
 
 cmd_append() {
-  local range="HEAD" key="$STORY_KEY" story="" bcps="" merged_at="" file="$ROOT/specs/metrics/cycle-times.yaml"
+  local range="HEAD" key="$STORY_KEY" story="" bcps="" bcp_plus="" merged_at="" file="$ROOT/specs/metrics/cycle-times.yaml"
   local telemetry=""  # path to harness telemetry JSON (GSD-2/gsd-pi format, optional)
   while [ $# -gt 0 ]; do
     case "$1" in
@@ -143,6 +143,8 @@ cmd_append() {
   agent_compression="null"; agent_tool_calls="null"; agent_api_requests="null"
   agent_duration_ms="null"; agent_model="null"; agent_tier="null"
   agent_downgraded="null"; agent_skills="[]"
+  agent_tokens_in="null"; agent_tokens_out="null"
+  agent_tokens_cache_read="null"; agent_tokens_cache_write="null"
 
   if [ -n "$telemetry" ] && [ -f "$telemetry" ]; then
     # GSD-2/gsd-pi snapshotUnitMetrics JSON format:

package/scripts/validate-agentic-ste.sh

@@ -79,7 +79,13 @@ ste_extract_prose() {
 
 ste_count_words() {
   # shellcheck disable=SC2207
-  local words=($1)
+  local words
+  # BUG-2026-08-05-ste-glob-wordcount: ($1) globs each word — a bare ** (markdown
+  # bold opener) expands to every file in cwd and inflates the count. Disable
+  # pathname expansion; safe here because this function only runs inside $( ).
+  set -f
+  words=($1)
+  set +f
   echo "${#words[@]}"
 }
 
@@ -168,6 +174,14 @@ name: fixture-long
 description: Long sentence fixture.
 ---
 Run the full local green stack including compliance verification gates sync skills trace stories and catalog drift check before any forward implementation work on this repository.
+EOF
+
+  cat > "$tmp/good-bold.md" <<'EOF'
+---
+name: fixture-bold
+description: Bold markdown list fixture (BUG-2026-08-05-ste-glob-wordcount).
+---
+1. **Check dependencies first.** DO inspect what your current dependencies already do before writing your own code.
 EOF
 
   local saved_strict=$STRICT saved_audit=$AUDIT
@@ -181,6 +195,13 @@ EOF
     return 1
   fi
 
+  VIOLATIONS=0
+  ste_audit_file "$tmp/good-bold.md"
+  if (( VIOLATIONS > 0 )); then
+    echo "SELF-TEST FAIL: bold-markdown fixture should pass (glob word-count bug, BUG-2026-08-05)" >&2
+    return 1
+  fi
+
   VIOLATIONS=0
   ste_audit_file "$tmp/bad-hedge.md"
   if (( VIOLATIONS == 0 )); then

package/scripts/validate-skill-yaml.py

@@ -1,5 +1,21 @@
 #!/usr/bin/env python3
-"""Validate YAML frontmatter in all .pi/skills/*/SKILL.md files."""
+"""Validate YAML frontmatter in SKILL.md files (sources + generated .pi/skills).
+
+Rules per file (sources and generated copies alike):
+
+  R1  file must start with the frontmatter delimiter '---' at byte 0.
+      pi (dist/utils/frontmatter.js) and the Agent Skills spec require the
+      opening delimiter to be the first characters of the file. Story tags
+      (`# story:` / `<!-- story: -->`) must live INSIDE the frontmatter block
+      as YAML comments, or below the closing delimiter — never above it,
+      or pi drops the skill as "description is required".
+      (story: BUG-2026-08-06-pi-drops-skills-frontmatter)
+  R2  frontmatter must parse as YAML (PyYAML or bundled simple_yaml).
+  R3  description must not embed 'model: sonnet' as literal text.
+
+Exits 1 on any failure. Called by the sync-skills.sh post-generation guard
+and by scripts/test-setup-no-pyyaml.sh.
+"""
 import glob
 import sys
 import os
@@ -17,64 +33,99 @@ except ModuleNotFoundError:
     from simple_yaml import parse_simple_yaml
     _load_yaml = parse_simple_yaml
 
-SKILLS_DIR = ".pi/skills"
-files = sorted(glob.glob(f"{SKILLS_DIR}/*/SKILL.md"))
+# Source of truth (skills/*/SKILL.md) and the pi-distributed copy (.pi/skills).
+SKILL_ROOTS = ["skills", ".pi/skills"]
+
+
+def validate_tree(root):
+    """Return (passed, delimiter_failures, parse_failures, semantic_failures)."""
+    files = sorted(glob.glob(os.path.join(root, "*/SKILL.md")))
+    parse_failures = []
+    semantic_failures = []
+    delimiter_failures = []
+    passed = []
+
+    for path in files:
+        skill_name = os.path.basename(os.path.dirname(path))
+        with open(path) as f:
+            content = f.read()
+
+        # R1 — pi's frontmatter gate: opening delimiter must be at byte 0.
+        if not content.startswith("---"):
+            delimiter_failures.append((skill_name, "file must start with '---' (story tags belong inside the frontmatter block or below it)"))
+            continue
 
-parse_failures = []
-semantic_failures = []
-passed = []
+        parts = content.split("---", 2)
+        if len(parts) < 3:
+            parse_failures.append((skill_name, "missing frontmatter delimiters"))
+            continue
 
-for path in files:
-    skill_name = os.path.basename(os.path.dirname(path))
-    with open(path) as f:
-        content = f.read()
+        frontmatter = parts[1]
 
-    parts = content.split("---", 2)
-    if len(parts) < 3:
-        parse_failures.append((skill_name, "missing frontmatter delimiters"))
-        continue
+        try:
+            data = _load_yaml(frontmatter)
+        except Exception as e:
+            parse_failures.append((skill_name, str(e).split("\n")[0]))
+            continue
 
-    frontmatter = parts[1]
+        if data is None:
+            parse_failures.append((skill_name, "empty frontmatter"))
+            continue
 
-    try:
-        data = _load_yaml(frontmatter)
-    except Exception as e:
-        parse_failures.append((skill_name, str(e).split("\n")[0]))
-        continue
+        # R3 — semantic check: model hint embedded in description
+        desc = data.get("description", "")
+        if desc and "model: sonnet" in desc:
+            semantic_failures.append(skill_name)
 
-    if data is None:
-        parse_failures.append((skill_name, "empty frontmatter"))
-        continue
+        passed.append(skill_name)
 
-    # Semantic check: model hint embedded in description
-    desc = data.get("description", "")
-    if desc and "model: sonnet" in desc:
-        semantic_failures.append(skill_name)
+    return passed, delimiter_failures, parse_failures, semantic_failures
 
-    passed.append(skill_name)
 
-# Report
-print(f"Total SKILL.md files: {len(files)}")
-print(f"Parse OK: {len(passed)}")
-print(f"Parse FAIL: {len(parse_failures)}")
-print(f"Semantic issues (model in desc): {len(semantic_failures)}")
+all_passed = []
+all_delimiter = []
+all_parse = []
+all_semantic = []
+
+for root in SKILL_ROOTS:
+    passed, delimiter_failures, parse_failures, semantic_failures = validate_tree(root)
+    all_passed += passed
+    all_delimiter += delimiter_failures
+    all_parse += parse_failures
+    all_semantic += semantic_failures
+    total = len(passed) + len(delimiter_failures) + len(parse_failures) + len(semantic_failures)
+    print(f"[{root}] SKILL.md files: {total} — parse OK: {len(passed)}")
+
+print()
+print(f"Total SKILL.md files: {len(all_passed) + len(all_delimiter) + len(all_parse) + len(all_semantic)}")
+print(f"Parse OK: {len(all_passed)}")
+print(f"Delimiter FAIL (must start with '---'): {len(all_delimiter)}")
+print(f"Parse FAIL: {len(all_parse)}")
+print(f"Semantic issues (model in desc): {len(all_semantic)}")
 print()
 
-if parse_failures:
+if all_delimiter:
+    print("=== DELIMITER FAILURES (file does not start with '---') ===")
+    for name, err in all_delimiter:
+        print(f"  {name}: {err}")
+    print()
+
+if all_parse:
     print("=== PARSE FAILURES ===")
-    for name, err in parse_failures:
+    for name, err in all_parse:
         print(f"  {name}: {err}")
     print()
 
-if semantic_failures:
+if all_semantic:
     print("=== SEMANTIC ISSUES (model embedded in description) ===")
-    for name in sorted(semantic_failures):
+    for name in sorted(all_semantic):
         print(f"  {name}")
     print()
 
-if parse_failures or semantic_failures:
-    print(f"FAIL: {len(parse_failures)} parse + {len(semantic_failures)} semantic issues")
+if all_delimiter or all_parse or all_semantic:
+    n = len(all_delimiter) + len(all_parse) + len(all_semantic)
+    print(f"FAIL: {n} issues ({len(all_delimiter)} delimiter + {len(all_parse)} parse + {len(all_semantic)} semantic)")
     sys.exit(1)
 else:
-    print("PASS: all SKILL.md YAML frontmatter valid")
+    print("PASS: all SKILL.md YAML frontmatter valid (sources + .pi/skills)")
     sys.exit(0)

package/skills/audit-code/SKILL.md

@@ -1,7 +1,7 @@
+---
 # story: e51s04
 # story: e45s18
 # story: e45s31
----
 name: audit-code
 model: haiku
 effort: standard

package/skills/craft-skill/SKILL.md

@@ -1,7 +1,7 @@
-# story: e45s02
-<!-- story: e45s12 -->
-<!-- story: e79s02 -->
 ---
+# story: e45s02
+# story: e45s12
+# story: e79s02
 name: craft-skill
 model: sonnet
 effort: standard

package/skills/deepen-architecture/SKILL.md

@@ -1,6 +1,6 @@
-# story: e45s31
-<!-- story: e45s14 -->
 ---
+# story: e45s31
+# story: e45s14
 name: deepen-architecture
 model: sonnet
 effort: standard

package/skills/deploy/SKILL.md

@@ -1,5 +1,5 @@
-<!-- story: e45s15 -->
 ---
+# story: e45s15
 name: deploy
 description: "Build → verify artifact → deploy → wait → smoke deployment pipeline. Platform-agnostic (MCP or CLI), with configurable timeout, retry with exponential backoff, and integrated health-check. The deploy half of CI/CD: run after build to push to production."
 model: sonnet

package/skills/develop-tdd/SKILL.md

@@ -1,10 +1,10 @@
+---
 # story: e80s01
 # story: e80s03
 # story: e51s04
 # story: e45s06
 # story: e45s08
 # story: e45s34
----
 name: develop-tdd
 model: sonnet
 effort: standard

package/skills/diagnose-stall/SKILL.md

@@ -1,5 +1,5 @@
-# story: e45s38
 ---
+# story: e45s38
 name: diagnose-stall
 model: haiku
 effort: light

package/skills/enforce-first/SKILL.md

@@ -1,5 +1,5 @@
-# story: e80s01
 ---
+# story: e80s01
 name: enforce-first
 model: haiku
 effort: standard

package/skills/fix-bug/SKILL.md

@@ -1,5 +1,5 @@
-# story: e51s04
 ---
+# story: e51s04
 name: fix-bug
 model: sonnet
 effort: standard

package/skills/gate-trace/SKILL.md

@@ -1,6 +1,6 @@
+---
 # story: e80s01
 # story: e45s05
----
 name: gate-trace
 description: "Deterministic traceability quality gate — reads coverage matrix + blind-spot data, applies decision rules with oracle confidence downgrade, emits PASS/CONCERNS/FAIL/WAIVED verdict. Use before release-branch to gate merges on traceability."
 # story: e38s06