wizz-method 2.25.3 → 2.26.0

package/package.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
   "name": "wizz-method",
-  "version": "2.25.3",
+  "version": "2.26.0",
   "description": "Wizz Method: método de agência orientado por IA em PT-BR",
   "keywords": [
     "agile",

package/src/skills-lib/wizz-cerebro/SKILL.md

@@ -19,7 +19,7 @@ description: >
 
 ```
 Session:   /iniciar  /salvar  /ver
-Routine:   /dia
+Routine:   /dia  /higiene
 Capture:   /dump  /decisao  /conteudo
 Research:  /prospect
 Memory:    /memoria
@@ -42,13 +42,14 @@ Read offset=(line-5) limit=15        # → reads only the needed excerpt
 | /salvar  | 2            | 3             |
 | /ver     | 1            | 0             |
 | /dia     | 1 per active project | 0    |
+| /higiene | 0            | only after the user confirms |
 | /dump    | 0            | 1             |
 | /decisao | 0            | 2             |
 | /memoria | 1            | 0             |
 
 **Per-file reading rules:**
 - `CEREBRO.md` — **never full Read**. Grep to find the project block (5–8 lines), Edit directly.
-- `projetos/[name].md` — Read the **last 80 lines** (tail). Sessions and current state sit at the end.
+- `projetos/[name].md` — Read the **top** (`limit=14`, the `Estado` block). Only `/salvar` also reads the last 40 lines (What's left + Sessions). Never a tail of 80.
 - `_decisions/`, `_content/` — Write directly, without reading the existing file.
 
 ## Vault
@@ -88,8 +89,9 @@ grep -m1 "^vault:" "/path/CEREBRO.md"
   _knowledge/
     about-me.md
     goals.md
-  _decisions/             ← YYYY-MM-DD-[slug].md, with frontmatter (date, project, topic,
-                            status, supersede). The `tema` field is what grep finds.
+  _decisions/             ← YYYY-MM-DD-[slug].md, with frontmatter (data, projeto, tema,
+                            status, supersede, substituida_por). The `tema` field is what grep finds.
+                            Schema and enums: see "Decision schema" below.
                             (`_learnings/` no longer exists: it was written to and never
                              read by any command, and was removed on 2026-08-28.
                              Stack gotchas and non-obvious bugs go to the agent's
@@ -100,14 +102,36 @@ grep -m1 "^vault:" "/path/CEREBRO.md"
     [name].md             ← current project state (structure: header + 12-line `Estado` block at the top + Preferences + Context + What's left + Sessions)
 ```
 
-**Linking rule:** always wiki-links `[[file]]` within the vault.
+**Linking rule:** `[[file]]` wiki-links only for notes that exist inside the vault. Anything outside it (auto-memory files, repo paths, skills) goes in backticks, never in `[[ ]]`: a wiki-link to a file outside the vault shows up as a broken link and pollutes the graph.
+
+## Decision schema (single, mandatory)
+
+Every file in `_decisions/` uses this frontmatter, with these exact Portuguese keys (never `date`/`project`):
+
+| key | rule |
+|---|---|
+| `data` | `YYYY-MM-DD` |
+| `projeto` | lowercase slug equal to the file name in `projetos/` (`meju`, not `MeJu`) |
+| `tema` | 2 to 4 comma-separated **concepts** (never words copied from the slug like `only`, `antes`, `fase1`); reuse terms that already exist, see `/decisao` step 2 |
+| `status` | one of `ativa`, `substituida`, `encerrada` (no other word) |
+| `supersede` | slug of the decision this one replaces (optional) |
+| `substituida_por` | slug of the decision that replaced this one (optional; the only name for this key) |
+
+The body must link its project (`Projeto: [[projetos/<slug>]]`) so the decision is never an orphan in the graph; when it replaces another, also link the old one.
+
+## Vault hygiene
+
+- A project file that grows past ~300 lines moves its oldest sessions to `_index/sessoes-<nome>.md` (move, never delete).
+- An archived project leaves a stub in `_archive/projetos/<nome>.md` so old links keep resolving.
+- No empty `Sem título*.canvas`, no `.DS_Store` committed.
+- Graph (`.obsidian/graph.json`, only touch it when the user asks): color by project, one group per project with the query `[projeto:<slug>] OR path:projetos/<slug>` (Tableau palette, so a decision and its project hub share a color), `file:CEREBRO` gold first, then one neutral group per folder. Keep labels for zoom only (`textFadeMultiplier` ~0.8, never negative), `nodeSizeMultiplier` ~1.5, `centerStrength` 0.5, `repelStrength` 10, `linkStrength` 1, `linkDistance` ~120. Every folder or project in a color query must exist.
 
 ## Commands (load on demand)
 
 Each command has its full step-by-step procedure in a reference file. Load only the file for the command group the user triggered:
 
 - `references/comandos-sessao.md` — `/iniciar` (set up new project), `/ver` (quick current state), `/salvar` (close session) with full steps, token budget and templates.
-- `references/comandos-rotina-e-captura.md` — `/dia` (daily briefing), `/dump` (quick capture), `/decisao` (record decision), `/conteudo` (content idea) with full steps and templates.
+- `references/comandos-rotina-e-captura.md` — `/dia` (daily briefing), `/higiene` (vault health report, then asks before fixing), `/dump` (quick capture), `/decisao` (record decision), `/conteudo` (content idea) with full steps and templates.
 - `references/comandos-pesquisa-e-memoria.md` — `/prospect` (research a prospect) and `/memoria` (read and feed the project's auto-memory from any platform) with full steps and templates.
 
 ## General rules
@@ -115,7 +139,7 @@ Each command has its full step-by-step procedure in a reference file. Load only
 - Wiki-links `[[file]]` within the vault.
 - Absolute dates in `YYYY-MM-DD` format, never relative ("yesterday", "last week"). Volatile facts (published version, release state, etc.) carry an `em YYYY-MM-DD:` prefix in the text; on re-reading, the agent treats it as a snapshot of that moment, not a current fact.
 - Never delete history — only append.
-- **Grep before writing.** Before creating a new decision, idea or record, search the topic in the destination folder. If the fact already exists, update the existing record; a decision that replaces another marks the old one as `superseded` instead of keeping both alive.
+- **Grep before writing.** Before creating a new decision, idea or record, search the topic in the destination folder. If the fact already exists, update the existing record; a decision that replaces another marks the old one as `substituida` instead of keeping both alive.
 - Session narrative lives in the project file: `Estado` block at the top (12 lines, overwritten on every `/salvar`) + 1 line in the `Sessões` table. `_index/sessions.md` no longer exists: it was written to and never read by any command, and was removed on 2026-08-28.
 - Summaries: 1 line per item.
 - CEREBRO.md should always stay under 150 useful lines — move long tables to `_index/`.

package/src/skills-lib/wizz-cerebro/references/comandos-rotina-e-captura.md

@@ -8,7 +8,7 @@ Load this file when running `/dia`, `/dump`, `/decisao` or `/conteudo`. Before a
 
 **Steps:**
 1. `grep -A4 "### " "$VAULT/CEREBRO.md"` → lists every project block (name + last session + pending)
-2. For each project with active status (🟢🟡): tail-read of 30 lines (`offset=n-30`)
+2. For each project with active status (🟢🟡): Read the top with `limit=14` (the `Estado` block); no tail
 3. Build a consolidated briefing
 4. Don't read `_knowledge/goals.md` unless the user explicitly asks
 
@@ -29,11 +29,45 @@ Load this file when running `/dia`, `/dump`, `/decisao` or `/conteudo`. Before a
 
 **Steps:**
 1. Receive the content (from the argument or ask in 1 message)
-2. Classify: idea / learning / decision / reference / content
+2. Classify: idea / decision / reference / content. A technical learning (stack gotcha, non-obvious bug) is not a vault fact: send it to the agent's auto-memory instead
 3. Write directly to the right file — without reading anything first
-4. `grep -n "## Conteúdo em aberto\|## Aprendizados" "$VAULT/CEREBRO.md"` → find the line for a 1-line Edit
+4. `grep -n "## Conteúdo em aberto" "$VAULT/CEREBRO.md"` → find the line for a 1-line Edit
 5. Confirm in 1 line
 
+## /higiene
+
+**Goal:** a health report of the vault, then an offer to fix. The check is read-only (zero Reads, bash only); **nothing is changed until the user confirms**.
+
+**Steps** (one bash call, `VAULT` from the Vault section):
+
+```bash
+cd "$VAULT"
+echo "decisão sem tema:";            grep -L "^tema:" _decisions/*.md
+echo "status fora do enum:";         grep -H "^status:" _decisions/*.md | grep -v -E "status: (ativa|substituida|encerrada)$"
+echo "decisão sem link de projeto:"; grep -L "\[\[projetos/" _decisions/*.md
+echo "projeto > 300 linhas:";        wc -l projetos/*.md | awk '$1>300 && $2!="total"'
+echo "CEREBRO.md linhas:";           wc -l < CEREBRO.md
+echo "pastas que não deviam existir:"; ls -d _learnings _moc 2>/dev/null
+echo "hubs (mais citados):";         grep -rhoE "\[\[[^]|#]+" --include="*.md" . | sed 's/\[\[//' | sort | uniq -c | sort -rn | head -10
+```
+
+For orphans: a note that no other file cites (`grep -rl "\[\[<name>" .` returns only itself) is an orphan; check only the notes listed above, not the whole vault.
+
+**Output:** one line per finding (`path · problem · suggested fix`), at most 20 lines, then 1 line with the totals. Nothing found = say so in 1 line and stop.
+
+**Then ask once** (AskUserQuestion where available, a numbered question otherwise) which groups to fix, multi-select, nothing pre-selected:
+
+| group | what the fix does | safe to batch |
+|---|---|---|
+| Decisions without project link | adds `Projeto: [[projetos/<projeto>]]` under the title, from the `projeto:` field | yes |
+| Decisions without `tema` or with a wrong `status` | fills `tema` from the existing vocabulary, normalizes `status` to the enum | yes |
+| Oversize projects | moves old sessions to `_index/sessoes-<nome>.md` (see `/salvar` step 6) | one project at a time |
+| Over-long `CEREBRO.md` | moves the oldest `Decisões recentes` to `_index/cerebro-archive.md` | yes |
+| Removed folders that came back (`_learnings`) | never deleted: lists the files and asks where each goes (auto-memory, `_decisions/`, `_archive/`) | no, one by one |
+| Orphans and hubs | informational only, no fix | n/a |
+
+Apply only the groups the user ticked, with surgical Edits (never rewrite a whole file), never delete anything (move only), and end with 1 line per group saying how many files changed. A group the user did not tick stays reported and untouched. If the question tool is not available and the user does not answer, change nothing.
+
 ## /decisao
 
 **Goal:** record a decision. 1 grep first, no Read.
@@ -47,28 +81,38 @@ grep -ril "<2 or 3 words about the topic>" "$VAULT/_decisions/" | head -5
 ```
 
 - Found a decision on the same topic: **don't create a new file**. Open the existing one and decide between two paths:
-  - the new decision **replaces** the old one → mark the old one with `status: superseded` and `supersedida_por: <new slug>`, and create the new one with `supersede: <old slug>`
+  - the new decision **replaces** the old one → mark the old one with `status: substituida` and `substituida_por: <new slug>`, and create the new one with `supersede: <old slug>`
   - the new decision **elaborates on** the old one → edit the old one, don't create a second
 - Found nothing: proceed to step 3.
 
-3. Write directly to `_decisions/YYYY-MM-DD-[slug].md` (template below)
-4. `grep -n "## Decisões recentes" "$VAULT/CEREBRO.md"` → line
-5. Edit to insert a wiki-link reference right below that line
-6. If there's a project: `grep -n "## Decisões" "$VAULT/projetos/$PROJ.md"` → Edit 1 line
+3. **Reuse the theme vocabulary** (1 bash call) before choosing `tema`:
+
+```bash
+grep -h "^tema:" "$VAULT/_decisions/"*.md | sed 's/^tema: *//' | tr ',' '\n' | sed 's/^ *//' | sort | uniq -c | sort -rn | head -40
+```
+
+Pick existing terms when they fit; add a new one only for a genuinely new concept. The theme names what the decision is about (`consentimento, tracking`), not the words of its title.
+
+4. Write directly to `_decisions/YYYY-MM-DD-[slug].md` (template below), with `Projeto: [[projetos/<slug>]]` under the title
+5. `grep -n "## Decisões recentes" "$VAULT/CEREBRO.md"` → line
+6. Edit to insert a wiki-link reference right below that line
+7. If there's a project: `grep -n "## Decisões" "$VAULT/projetos/$PROJ.md"` → Edit 1 line
 
 **Template `_decisions/YYYY-MM-DD-[slug].md`:**
 ```markdown
 ---
 data: YYYY-MM-DD
-projeto: [name]
+projeto: [lowercase slug equal to the file name in projetos/]
 tema: [2 to 4 keywords separated by comma, this is what grep finds]
-status: ativa
+status: ativa   # ativa | substituida | encerrada
 supersede: [slug of the decision this replaces, or omit]
-supersedida_por: [filled in later, when another decision replaces this one]
+substituida_por: [filled in later, when another decision replaces this one]
 ---
 
 # [Title]
 
+Projeto: [[projetos/<slug>]]
+
 ## Decision
 [1-2 lines]
 

package/src/skills-lib/wizz-cerebro/references/comandos-sessao.md

@@ -84,6 +84,14 @@ What does **not** go in the block: history, decisions, learnings, a long list of
 1. Read `projetos/[name].md` with `limit=14` (the State block is at the top)
 2. Return the block almost as-is, without rewriting it
 
+**`/ver decisoes <tema>`:** read the decisions about a theme, without opening files. One bash call:
+
+```bash
+grep -l "^tema:.*<tema>" "$VAULT/_decisions/"*.md | xargs grep -L "^status: substituida" | sort -r | head -5
+```
+
+Return at most 5 lines `date · title · status` (title = the first `# ` line). Open one decision only if the user asks. Decisions marked `substituida` are skipped on purpose: the one that replaced them is the live one.
+
 **Old file, with no State block at the top:** falls onto the slow path exactly once — `wc -l` and Read of the last 60 lines — and when returning it, offer to create the block. Once accepted, the next `/ver` goes back to costing 1 short Read.
 
 ## /salvar
@@ -143,4 +151,10 @@ grep -c "^| " "$VAULT/projetos/$PROJ.md"
 
 - Sessions table with more than 15 lines: move the oldest ones to `_index/cerebro-archive.md`, under `## Sessões antigas — [project]`, leaving the 15 newest in the project file. Never delete, only move.
 
+```bash
+wc -l "$VAULT/projetos/$PROJ.md"
+```
+
+- Project file over ~300 lines: move the oldest sessions (and any history block that is not `Estado`, `O que falta` or the 15 newest sessions) to `_index/sessoes-$PROJ.md`, leaving a 1-line `[[_index/sessoes-$PROJ]]` pointer. Never delete, only move.
+
 **7. Confirm in 2 lines** what was saved.