feat(observabilidade): tabela worklog_comments + parser HTML + importer MCP

- Schema worklog_comments (id, discussion, parent, datas, staff, campos parseados em JSON)
- Parser HTML tolerante (h2/h3/h4) extrai title, task_ref, duration, work_items,
  files_modified, problems, patterns_text, actions
- Módulo worklog-import com paginação MCP get_discussion_comments
- Helper mcp-client.ts partilhado (gateway MCP JSON-RPC + SSE)
- Dep runtime: node-html-parser

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-04-23 03:07:09 +01:00
parent 86770b1570
commit 11f9833aac
5 changed files with 685 additions and 0 deletions
+131
View File
@@ -27,6 +27,31 @@ export interface PatternRecord {
consecutive_weeks: number
}
export interface WorklogCommentRecord {
id: number
discussion_id: number
created_at: string
staff_id: number | null
title: string | null
task_ref: string | null
duration_sec: number | null
work_items: string[]
files_modified: string[]
problems: { problema: string; solucao: string }[]
patterns_text: string[]
actions: { tipo: string; descricao: string; prioridade: string | null }[]
raw_html: string
imported_at: string
}
export interface WorklogFilters {
discussion_id?: number
task_ref?: string
sinceIso?: string
limit?: number
offset?: number
}
export interface SessionsDb {
upsertSession(meta: SessionMeta): void
upsertMany(metas: SessionMeta[]): void
@@ -37,6 +62,10 @@ export interface SessionsDb {
upsertPattern(p: PatternRecord): void
getPatternsByWeek(week: string): PatternRecord[]
getConsecutiveWeeks(pattern_key: string, uptoWeek: string): number
upsertWorklogComment(c: WorklogCommentRecord): { inserted: boolean }
hasWorklogComment(id: number): boolean
listWorklogComments(filters: WorklogFilters): WorklogCommentRecord[]
countWorklogComments(filters?: WorklogFilters): number
rawDb(): Database.Database
close(): void
}
@@ -81,6 +110,25 @@ CREATE TABLE IF NOT EXISTS patterns (
);
CREATE INDEX IF NOT EXISTS idx_patterns_week ON patterns(week_iso);
CREATE INDEX IF NOT EXISTS idx_patterns_key ON patterns(pattern_key);
CREATE TABLE IF NOT EXISTS worklog_comments (
id INTEGER PRIMARY KEY,
discussion_id INTEGER NOT NULL,
created_at TEXT NOT NULL,
staff_id INTEGER,
title TEXT,
task_ref TEXT,
duration_sec INTEGER,
work_items TEXT NOT NULL,
files_modified TEXT NOT NULL,
problems_json TEXT NOT NULL,
patterns_text TEXT NOT NULL,
actions_json TEXT NOT NULL,
raw_html TEXT NOT NULL,
imported_at TEXT NOT NULL
);
CREATE INDEX IF NOT EXISTS idx_wc_discussion ON worklog_comments(discussion_id, created_at DESC);
CREATE INDEX IF NOT EXISTS idx_wc_task ON worklog_comments(task_ref);
`
function rowToMeta(row: Record<string, unknown>): SessionMeta {
@@ -272,6 +320,89 @@ export function openSessionsDb(dbPath: string): SessionsDb {
}
return count
},
upsertWorklogComment(c: WorklogCommentRecord): { inserted: boolean } {
const existing = db.prepare('SELECT 1 FROM worklog_comments WHERE id = ?').get(c.id)
const inserted = !existing
db.prepare(`
INSERT INTO worklog_comments (id, discussion_id, created_at, staff_id, title, task_ref,
duration_sec, work_items, files_modified, problems_json, patterns_text, actions_json,
raw_html, imported_at)
VALUES (@id, @discussion_id, @created_at, @staff_id, @title, @task_ref,
@duration_sec, @work_items, @files_modified, @problems_json, @patterns_text, @actions_json,
@raw_html, @imported_at)
ON CONFLICT(id) DO UPDATE SET
discussion_id = excluded.discussion_id,
created_at = excluded.created_at,
staff_id = excluded.staff_id,
title = excluded.title,
task_ref = excluded.task_ref,
duration_sec = excluded.duration_sec,
work_items = excluded.work_items,
files_modified = excluded.files_modified,
problems_json = excluded.problems_json,
patterns_text = excluded.patterns_text,
actions_json = excluded.actions_json,
raw_html = excluded.raw_html,
imported_at = excluded.imported_at
`).run({
id: c.id,
discussion_id: c.discussion_id,
created_at: c.created_at,
staff_id: c.staff_id,
title: c.title,
task_ref: c.task_ref,
duration_sec: c.duration_sec,
work_items: JSON.stringify(c.work_items),
files_modified: JSON.stringify(c.files_modified),
problems_json: JSON.stringify(c.problems),
patterns_text: JSON.stringify(c.patterns_text),
actions_json: JSON.stringify(c.actions),
raw_html: c.raw_html,
imported_at: c.imported_at,
})
return { inserted }
},
hasWorklogComment(id: number): boolean {
return !!db.prepare('SELECT 1 FROM worklog_comments WHERE id = ?').get(id)
},
listWorklogComments(filters: WorklogFilters): WorklogCommentRecord[] {
const parts: string[] = []
const params: Record<string, unknown> = {}
if (filters.discussion_id) { parts.push('discussion_id = @discussion_id'); params.discussion_id = filters.discussion_id }
if (filters.task_ref) { parts.push('task_ref = @task_ref'); params.task_ref = filters.task_ref }
if (filters.sinceIso) { parts.push('created_at >= @since'); params.since = filters.sinceIso }
const where = parts.length ? 'WHERE ' + parts.join(' AND ') : ''
const limit = filters.limit ?? 1000
const offset = filters.offset ?? 0
const rows = db.prepare(`SELECT * FROM worklog_comments ${where} ORDER BY created_at DESC LIMIT @limit OFFSET @offset`)
.all({ ...params, limit, offset }) as Record<string, unknown>[]
return rows.map((r) => ({
id: r.id as number,
discussion_id: r.discussion_id as number,
created_at: r.created_at as string,
staff_id: (r.staff_id as number | null) ?? null,
title: (r.title as string | null) ?? null,
task_ref: (r.task_ref as string | null) ?? null,
duration_sec: (r.duration_sec as number | null) ?? null,
work_items: JSON.parse(r.work_items as string),
files_modified: JSON.parse(r.files_modified as string),
problems: JSON.parse(r.problems_json as string),
patterns_text: JSON.parse(r.patterns_text as string),
actions: JSON.parse(r.actions_json as string),
raw_html: r.raw_html as string,
imported_at: r.imported_at as string,
}))
},
countWorklogComments(filters?: WorklogFilters): number {
const parts: string[] = []
const params: Record<string, unknown> = {}
if (filters?.discussion_id) { parts.push('discussion_id = @discussion_id'); params.discussion_id = filters.discussion_id }
if (filters?.task_ref) { parts.push('task_ref = @task_ref'); params.task_ref = filters.task_ref }
if (filters?.sinceIso) { parts.push('created_at >= @since'); params.since = filters.sinceIso }
const where = parts.length ? 'WHERE ' + parts.join(' AND ') : ''
const row = db.prepare(`SELECT COUNT(*) as c FROM worklog_comments ${where}`).get(params) as { c: number }
return row.c
},
rawDb(): Database.Database {
return db
},
+69
View File
@@ -0,0 +1,69 @@
/**
* Cliente HTTP mínimo para o gateway MCP (JSON-RPC 2.0 sobre HTTP).
*
* Suporta resposta em JSON puro ou SSE (text/event-stream). Partilhado entre
* os scripts de Observabilidade (patterns + worklog import).
*/
export interface MCPToolCallResult {
content?: Array<{ type: string; text: string }>
isError?: boolean
}
export async function callMcpTool(
tool: string,
args: Record<string, unknown>,
): Promise<MCPToolCallResult> {
const url = process.env.MCP_GATEWAY_URL ?? 'https://gateway.descomplicar.pt/v1/desk-crm/mcp'
const token = process.env.MCP_GATEWAY_TOKEN
if (!token) throw new Error('MCP_GATEWAY_TOKEN não definido')
const body = {
jsonrpc: '2.0',
id: Date.now(),
method: 'tools/call',
params: { name: tool, arguments: args },
}
const res = await fetch(url, {
method: 'POST',
headers: {
Authorization: `Bearer ${token}`,
'Content-Type': 'application/json',
Accept: 'application/json, text/event-stream',
},
body: JSON.stringify(body),
})
if (!res.ok) {
const txt = await res.text().catch(() => '')
throw new Error(`MCP gateway ${res.status}: ${txt.slice(0, 300)}`)
}
const raw = await res.text()
let payload: string | null = null
for (const line of raw.split(/\r?\n/)) {
const trimmed = line.trim()
if (!trimmed) continue
if (trimmed.startsWith('data: ')) {
payload = trimmed.slice(6)
break
}
if (trimmed.startsWith('{')) {
payload = trimmed
break
}
}
if (!payload) throw new Error(`MCP resposta sem payload JSON: ${raw.slice(0, 200)}`)
const parsed = JSON.parse(payload) as { error?: unknown; result?: MCPToolCallResult }
if (parsed.error) throw new Error(`MCP error: ${JSON.stringify(parsed.error)}`)
const result = parsed.result as MCPToolCallResult | undefined
if (result?.isError) {
const txt = result.content?.map((c) => c.text).join('\n') ?? ''
throw new Error(`MCP tool ${tool} devolveu isError: ${txt.slice(0, 300)}`)
}
return result ?? {}
}
/** Extrai o primeiro bloco de texto JSON-encoded do resultado MCP. */
export function extractMcpJsonPayload<T = unknown>(r: MCPToolCallResult): T {
const text = r.content?.find((c) => c.type === 'text')?.text
if (!text) throw new Error('MCP result sem content text')
return JSON.parse(text) as T
}
+351
View File
@@ -0,0 +1,351 @@
/**
* Importer dos comentários das discussões Desk #31 (Logs), #32 (Reflexões)
* e #33 (Acções de Melhoria) para a tabela `worklog_comments`.
*
* Parser HTML tolerante — aceita ambos formatos produzidos pelo skill
* `gestao:worklog` (versão antiga usava `<h2>/<h3>` inline-styled, versão
* nova usa `<h4>` limpos). Secções identificadas por título normalizado
* (ex.: "trabalho realizado", "ficheiros modificados", "problemas",
* "padrões detectados", "acções sugeridas").
*/
import { parse, type HTMLElement } from 'node-html-parser'
import type { SessionsDb, WorklogCommentRecord } from './db.js'
import { callMcpTool, extractMcpJsonPayload } from './mcp-client.js'
export interface ParsedWorklogComment {
id: number
discussion_id: number
created_at: string
staff_id: number | null
title: string | null
task_ref: string | null
duration_sec: number | null
work_items: string[]
files_modified: string[]
problems: { problema: string; solucao: string }[]
patterns_text: string[]
actions: { tipo: string; descricao: string; prioridade: string | null }[]
raw_html: string
}
interface RawComment {
id: number
discussion_id: number
content: string
created: unknown
staff_id: number | null
children?: RawComment[]
}
/** Remove whitespace redundante. */
function norm(s: string): string {
return s.replace(/\s+/g, ' ').trim()
}
/** Converte string livre para chave de secção (lowercase, sem acentos, sem pontuação). */
function sectionKey(s: string): string {
return s
.toLowerCase()
.normalize('NFD')
.replace(/[̀-ͯ]/g, '')
.replace(/[^a-z0-9 ]/g, ' ')
.replace(/\s+/g, ' ')
.trim()
}
const SECTION_WORK = new Set(['trabalho realizado', 'o que foi feito', 'feito', 'realizado', 'trabalho'])
const SECTION_FILES = new Set(['ficheiros modificados', 'ficheiros alterados', 'files modified', 'ficheiros'])
const SECTION_PROBLEMS = new Set(['problemas solucoes', 'problemas', 'solucoes', 'problemas e solucoes', 'problemas solucao'])
const SECTION_PATTERNS = new Set(['padroes detectados', 'padroes', 'patterns', 'insights'])
const SECTION_ACTIONS = new Set(['accoes sugeridas', 'accoes', 'acoes sugeridas', 'acoes', 'actions', 'accoes de melhoria'])
/** Extrai data ISO do título (YYYY-MM-DD [HH:MM]) ou devolve null. */
function parseDateFromTitle(title: string): string | null {
const m = title.match(/(\d{4})-(\d{2})-(\d{2})(?:[ T](\d{2}):(\d{2}))?/)
if (!m) return null
const [, y, mo, d, hh, mm] = m
if (hh && mm) return `${y}-${mo}-${d}T${hh}:${mm}:00Z`
return `${y}-${mo}-${d}T00:00:00Z`
}
/** Tenta extrair "Tarefa: #ID" ou similar. */
function parseTaskRef(text: string): string | null {
const m = text.match(/(?:Tarefa|Task|Ticket)[:\s]*(#?\d+)/i)
if (m) return m[1].startsWith('#') ? m[1] : `#${m[1]}`
const bare = text.match(/#(\d{3,6})/)
return bare ? `#${bare[1]}` : null
}
/** "~2h 30m" / "~45 min" / "5 minutos" → segundos. */
function parseDuration(text: string): number | null {
const m = text.match(/~?\s*(\d+)\s*h\s*(\d+)?\s*m?/i)
if (m) {
const h = parseInt(m[1], 10)
const mm = m[2] ? parseInt(m[2], 10) : 0
return h * 3600 + mm * 60
}
const mm = text.match(/~?\s*(\d+)\s*(?:min|minutos|m)/i)
if (mm) return parseInt(mm[1], 10) * 60
return null
}
/** Extrai texto de um elemento, incluindo inner HTML como plain text. */
function textOf(el: HTMLElement): string {
return norm(el.text ?? '')
}
/** Colecta items de uma UL ou lista no mesmo nível que vem depois de um cabeçalho. */
function collectFollowingListItems(heading: HTMLElement): string[] {
const items: string[] = []
let cur: HTMLElement | null = heading.nextElementSibling
while (cur) {
const tag = cur.rawTagName?.toLowerCase()
if (tag && /^h[1-6]$/.test(tag)) break
if (tag === 'ul' || tag === 'ol') {
for (const li of cur.querySelectorAll('li')) {
const t = textOf(li)
if (t) items.push(t)
}
} else if (tag === 'p') {
// Alguns comentários partem o UL em múltiplos <p>; vasculha <li> dentro
for (const li of cur.querySelectorAll('li')) {
const t = textOf(li)
if (t) items.push(t)
}
}
cur = cur.nextElementSibling
}
return items
}
/** Parse item "[Tipo] descrição" ou "Tipo: descrição (Px)". */
function parseActionItem(raw: string): { tipo: string; descricao: string; prioridade: string | null } {
// Remove checkbox inicial "[ ]" ou "[x]" se existir
let s = raw.trim().replace(/^\[[\s xX✓]\]\s*/, '')
const bracket = s.match(/^\[([^\]]+)\]\s*(.+)$/)
let tipo = 'Geral'
let rest = s
if (bracket) {
tipo = bracket[1].trim()
rest = bracket[2].trim()
}
const prio = rest.match(/\b(P[0-4])\b/i)
return {
tipo,
descricao: rest,
prioridade: prio ? prio[1].toUpperCase() : null,
}
}
/** Parse problema/solução. Heurística: "Problema: X | Solução: Y" ou pares de <li>. */
function parseProblemItem(raw: string): { problema: string; solucao: string } {
const s = raw.trim()
const split = s.split(/\s*(?:->|→|\|\s*Solu[çc][ãa]o:|\s*Solu[çc][ãa]o:)\s*/i)
if (split.length >= 2) {
return {
problema: split[0].replace(/^Problema:\s*/i, '').trim(),
solucao: split.slice(1).join(' ').trim(),
}
}
return { problema: s, solucao: '' }
}
/** Extrai lista "bruta" de todas as <li> dentro do HTML (fallback). */
function extractAllLiItems(root: HTMLElement): string[] {
return root
.querySelectorAll('li')
.map((li) => textOf(li))
.filter(Boolean)
}
export function parseWorklogHtml(
html: string,
meta: { id: number; discussion_id: number; created_at: string; staff_id?: number | null },
): ParsedWorklogComment {
const root = parse(html || '')
const headings = root.querySelectorAll('h1, h2, h3, h4, h5, h6')
// Título: primeiro heading não vazio
let title: string | null = null
for (const h of headings) {
const t = textOf(h)
if (t) { title = t; break }
}
// Data: preferir `meta.created_at` se válido; senão extrair do título ou do texto
let createdAt = meta.created_at
if (!createdAt || createdAt === '1970-01-01T00:00:00.000Z' || createdAt.startsWith('1970')) {
const fromTitle = title ? parseDateFromTitle(title) : null
if (fromTitle) createdAt = fromTitle
else {
const fromText = parseDateFromTitle(textOf(root).slice(0, 500))
createdAt = fromText ?? new Date().toISOString()
}
}
const fullText = textOf(root)
const taskRef = parseTaskRef(fullText)
const durationSec = parseDuration(fullText)
// Indexa secções por chave normalizada
const sections = new Map<string, HTMLElement>()
for (const h of headings) {
const key = sectionKey(textOf(h))
if (!sections.has(key)) sections.set(key, h)
}
function findSection(target: Set<string>): HTMLElement | null {
for (const [k, el] of sections) {
if (target.has(k)) return el
}
// match parcial (ex.: "trabalho realizado manutenção" — começa com)
for (const [k, el] of sections) {
for (const t of target) {
if (k.startsWith(t) || t.startsWith(k)) return el
}
}
return null
}
const workHeading = findSection(SECTION_WORK)
const filesHeading = findSection(SECTION_FILES)
const problemsHeading = findSection(SECTION_PROBLEMS)
const patternsHeading = findSection(SECTION_PATTERNS)
const actionsHeading = findSection(SECTION_ACTIONS)
const workItems = workHeading ? collectFollowingListItems(workHeading) : []
const filesModified = filesHeading ? collectFollowingListItems(filesHeading) : []
const problemsRaw = problemsHeading ? collectFollowingListItems(problemsHeading) : []
const patternsText = patternsHeading ? collectFollowingListItems(patternsHeading) : []
const actionsRaw = actionsHeading ? collectFollowingListItems(actionsHeading) : []
// Fallback: se nenhuma secção encontrada mas existem <li>, e a discussão é #33,
// tratar tudo como acções (formato diferente das outras discussões)
let actions = actionsRaw.map(parseActionItem)
if (meta.discussion_id === 33 && actions.length === 0) {
actions = extractAllLiItems(root).map(parseActionItem)
}
const problems = problemsRaw.map(parseProblemItem)
return {
id: meta.id,
discussion_id: meta.discussion_id,
created_at: createdAt,
staff_id: meta.staff_id ?? null,
title,
task_ref: taskRef,
duration_sec: durationSec,
work_items: workItems,
files_modified: filesModified,
problems,
patterns_text: patternsText,
actions,
raw_html: html,
}
}
/** Converte o campo `created` devolvido pelo Desk MCP (pode ser objecto vazio). */
function normalizeMcpDate(v: unknown): string {
if (!v) return ''
if (typeof v === 'string') return v
if (typeof v === 'object') {
const obj = v as Record<string, unknown>
if (typeof obj.date === 'string') return obj.date
if (typeof obj.datetime === 'string') return obj.datetime
}
return ''
}
/** Achata a árvore de comentários (comentários com children recursivos). */
function flattenComments(comments: RawComment[]): RawComment[] {
const out: RawComment[] = []
for (const c of comments) {
out.push(c)
if (c.children && c.children.length) {
out.push(...flattenComments(c.children))
}
}
return out
}
export interface ImportResult {
discussion_id: number
fetched: number
imported: number
updated: number
skipped: number
errors: number
}
/**
* Importa todos os comentários de uma discussão Desk. Paginação por `limit`/`offset`.
* Idempotente por `id` — comentários já existentes sofrem update (raw_html pode mudar).
*/
export async function importWorklogDiscussion(
db: SessionsDb,
discussionId: number,
opts: { sinceIso?: string; pageSize?: number; maxPages?: number } = {},
): Promise<ImportResult> {
// O MCP desk-crm parece clampar resultados em 200/página independentemente do limit.
// Pedimos 200 e iteramos offset até a resposta vir vazia.
const pageSize = opts.pageSize ?? 200
const maxPages = opts.maxPages ?? 20
const result: ImportResult = {
discussion_id: discussionId,
fetched: 0,
imported: 0,
updated: 0,
skipped: 0,
errors: 0,
}
let offset = 0
for (let page = 0; page < maxPages; page++) {
const raw = await callMcpTool('get_discussion_comments', {
discussion_id: discussionId,
limit: pageSize,
offset,
})
const payload = extractMcpJsonPayload<{
success?: boolean
comments?: RawComment[]
}>(raw)
const pageComments = flattenComments(payload.comments ?? [])
if (pageComments.length === 0) break
result.fetched += pageComments.length
const importedAt = new Date().toISOString()
for (const c of pageComments) {
try {
const createdStr = normalizeMcpDate(c.created)
const parsed = parseWorklogHtml(c.content ?? '', {
id: c.id,
discussion_id: c.discussion_id ?? discussionId,
created_at: createdStr || '',
staff_id: c.staff_id,
})
if (opts.sinceIso && parsed.created_at < opts.sinceIso) {
result.skipped++
continue
}
const record: WorklogCommentRecord = {
...parsed,
imported_at: importedAt,
}
const { inserted } = db.upsertWorklogComment(record)
if (inserted) result.imported++
else result.updated++
} catch (e) {
console.error(`[worklog-import] erro a parsear comentário #${c.id}:`, (e as Error).message)
result.errors++
}
}
// Avança offset; quando próxima página vier vazia, o while quebra na próxima iter.
offset += pageComments.length
// Safety: se MCP devolveu 0, para
if (pageComments.length === 0) break
}
return result
}