Files

T

ealmeida 6285be6c2e feat(gestao): adicionar 9 skills /clip-* + migrar 5 para diag tools MCP

Skills clip-* nunca tinham sido committed. Adicionadas todas (9):
clip, clip-agent, clip-health, clip-instructions, clip-issue, clip-org,
clip-routine, clip-skill, clip-vision.

Migração para mcp__paperclip__diag_* (17 substituições em 5 skills):
- clip: 5 substituições (agents_by_status, false_blockers, token burn,
  stuck routines, company_skills_summary)
- clip-agent: 2 (agent_full_context consolida 4 passos, false_blockers)
- clip-health: 8 (budget_orphans, missing_permissions, missing_heartbeat,
  routine_triggers_broken, false_blockers, heartbeat_token_usage,
  prompt_too_long_errors, stuck_routines, zombie_parents)
- clip-org: 1 (agent_hierarchy)
- clip-routine: 1 (routine_triggers_broken)

Sem substituições (CRUD-específico sem diag_* equivalente):
clip-instructions, clip-issue, clip-skill — mantêm psql.

Refs: Desk #2041, mcp-paperclip feature/diagnostics-db

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-04-07 03:59:54 +01:00

4.8 KiB

Raw Blame History

name, description, context

name	description	context
clip	Dashboard rápido Paperclip — estado agentes, issues abertas, heartbeats recentes, alertas. Usar quando "clip", "paperclip", "estado clip", "agentes clip".	fork

/clip — Dashboard Paperclip

Dashboard rapido do estado do Clip (Paperclip Orquestrador Descomplicar).

Constantes

BD: PGPASSWORD="paperclip" psql -h localhost -p 54329 -U paperclip -d paperclip
COMPANY_ID: ebe10308-efd7-453f-86ab-13e6fe84004f

Procedimento

Executar os 5 passos em paralelo (silenciosamente), depois apresentar dashboard compacto.

Passo 1: Estado do servico

systemctl --user status paperclip 2>&1 | head -5

Passo 2: Agentes por status

Invocar tool MCP: mcp__paperclip__diag_agents_by_status

Passo 3: Issues abertas

SELECT i.title, i.status, i.priority, a.name as assignee
FROM issues i LEFT JOIN agents a ON i.assignee_agent_id = a.id
WHERE i.company_id = 'ebe10308-efd7-453f-86ab-13e6fe84004f'
AND i.status NOT IN ('done','cancelled')
ORDER BY
  CASE i.priority WHEN 'critical' THEN 1 WHEN 'high' THEN 2 WHEN 'medium' THEN 3 ELSE 4 END,
  i.status;

Passo 4: Ultimos 10 heartbeats

SELECT a.name, hr.status, hr.started_at, hr.finished_at
FROM heartbeat_runs hr JOIN agents a ON hr.agent_id = a.id
WHERE a.company_id = 'ebe10308-efd7-453f-86ab-13e6fe84004f'
ORDER BY hr.started_at DESC LIMIT 10;

Passo 5: Alertas

Heartbeats falhados nas ultimas 24h:

SELECT a.name, LEFT(hr.error, 80) as erro, hr.started_at
FROM heartbeat_runs hr JOIN agents a ON hr.agent_id = a.id
WHERE a.company_id = 'ebe10308-efd7-453f-86ab-13e6fe84004f'
AND hr.status = 'failed'
AND hr.started_at > NOW() - INTERVAL '24 hours'
ORDER BY hr.started_at DESC;

Falsos blockers (INC-07):

Invocar tool MCP: mcp__paperclip__diag_false_blockers

Se resultado > 0 → listar como alerta: "N issues blocked com sub-tasks activas — provavelmente deveriam ser in_progress".

Token burn — agentes com consumo excessivo de tokens nas últimas 24h (INC-12):

Invocar tool MCP: mcp__paperclip__diag_heartbeat_token_usage(hours=24)

Se agentes_risco > 0 → alertar: "TOKEN BURN: N agentes com runs >500K cache tokens — usar /clip-health check 19 para detalhes".

Routines presas (INC-08) — a ligação issue→routine está em routine_runs.linked_issue_id:

Invocar tool MCP: mcp__paperclip__diag_stuck_routines(hours=4)

Se routines_presas > 0 → alertar: "ROUTINE BLOQUEADA: N issues de routine paradas >4h — usar /clip-health check 20".

Budget (agentes com >0 configurado):

SELECT name, budget_monthly_cents, spent_monthly_cents,
  CASE WHEN budget_monthly_cents > 0
    THEN ROUND(spent_monthly_cents::numeric / budget_monthly_cents * 100, 1)
    ELSE 0 END as pct
FROM agents
WHERE company_id = 'ebe10308-efd7-453f-86ab-13e6fe84004f'
AND budget_monthly_cents > 0
ORDER BY pct DESC;

Passo 6: Saude API (endpoint dashboard)

Verificar se os endpoints criticos respondem:

HEALTH=$(curl -s -o /dev/null -w "%{http_code}" http://localhost:3100/api/health)
DASH=$(curl -s -o /dev/null -w "%{http_code}" http://localhost:3100/api/companies/ebe10308-efd7-453f-86ab-13e6fe84004f/dashboard)
BADGES=$(curl -s -o /dev/null -w "%{http_code}" http://localhost:3100/api/companies/ebe10308-efd7-453f-86ab-13e6fe84004f/sidebar-badges)
echo "health=$HEALTH dashboard=$DASH badges=$BADGES"

health=200 = OK (sem auth)
dashboard/badges=401 = OK (auth funciona, endpoint existe)
dashboard/badges=404 ou 500 = CRITICO (endpoint em falha, provavelmente refs orfas na BD)

Verificar instancias duplicadas:

ss -tlnp | grep -E "310[0-9]" | wc -l

1 = OK
2+ = AVISO (processos orfaos, recomendar limpeza)

Passo 7: Company skills

Invocar tool MCP: mcp__paperclip__diag_company_skills_summary

Formato de output

Apresentar como dashboard compacto:

## Clip Dashboard — [data/hora]

**Servico:** [running/stopped] ha [tempo]
**Agentes:** [N] total — [n] running, [n] idle, [n] paused
**API:** health=[code] dashboard=[code] badges=[code] | Instancias: [N]
**Skills:** [N] instaladas ([n] locais, [n] skills.sh, [n] github)

### Issues abertas
| Titulo | Status | Prioridade | Assignee |
...

### Heartbeats recentes
| Agente | Status | Inicio | Fim |
...

### Alertas
- [CRITICO] descricao
- [TOKEN BURN] N agentes >500K tokens — /clip-health check 19
- [ROUTINE BLOQUEADA] N issues paradas >4h — /clip-health check 20

Referencias

Manual: Hub 06-Operacoes/Documentacao/Manuais/Paperclip/
API: 04-Stack/02.06-Clip/skills.md
Spec: 04-Stack/02.06-Clip/docs/superpowers/specs/2026-03-28-clip-grande-visao-design.md

Healing Log

Registo de erros conhecidos e como evitá-los. Lido automaticamente antes de executar.

{"date":"","issue":"","fix":"","source":"user|auto"}

Adicionar nova linha após cada erro corrigido.

4.8 KiB Raw Blame History