com um clique
budget-gatekeeper
Token and cost tracking with model tier enforcement
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Menu
Token and cost tracking with model tier enforcement
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Demo pipeline state machine — 7-phase autonomous sequence with delegation context templates, phase transition logic, BLOCKED recovery strategies, and demo-state.json schema. Used exclusively by demo-conductor.
Cinematic narration style guide for demo-conductor — ANSI-coloured banner formats, live pipeline scoreboard, audience-facing language, phase summaries, and error narration patterns. Keeps the demo presentation-quality throughout.
Standardized completion and escalation protocol for subagent responses. Ensures the conductor can machine-parse every subagent return. Use when reporting completion status back to the orchestrator.
Cross-session learnings lifecycle — schema, storage, retrieval, and pruning of lessons learned during orchestrator sessions. Use when managing learnings via the /learn command.
Agent Teams assembly and task injection — selects appropriate team, validates prerequisites, estimates cost, injects tasks into the shared task list, and manages team lifecycle.
Complexity-based task routing with model tier selection
| name | budget-gatekeeper |
| description | Token and cost tracking with model tier enforcement |
| argument-hint | <session-context> |
| user-invocable | false |
Track usage by tier to monitor cost distribution:
| Tier | Model | Relative Cost | Target Budget Share |
|---|---|---|---|
| heavy | Opus 4.6 | 1.0x (baseline) | ≤ 25% of session tokens |
| default | Sonnet 4.6 | ~0.2x | 60-70% of session tokens |
| fast | Haiku 4.5 | ~0.04x | 5-15% of session tokens |
Subagent delegations are automatically logged to artifacts/sessions/delegation-log.jsonl by the SubagentStart and SubagentStop hooks. At each delegation, record:
| Metric | Warning | Hard Limit |
|---|---|---|
| Total delegations | 8 | 15 |
| Heavy-tier delegations | 4 | 8 |
| Estimated session cost | $2.00 | $5.00 |
| Single agent turns | 30 | 60 |
When ORCH_TEAMS_ENABLED=true, additional limits apply:
| Metric | Warning | Hard Limit |
|---|---|---|
| Active teams per session | — | 1 (CC runtime enforces) |
| Tasks per team | 10 | ORCH_TEAM_MAX_TASKS (default 20) |
| Estimated team session cost | $10.00 | User-confirmed before assembly |
| Teammates per team | — | ORCH_TEAM_SIZE_MAX (default 3) |
Team sessions cost approximately 7x a single session. The team-routing skill presents a cost estimate and requires explicit user confirmation (y) before assembling any team. Log team task completions from delegation-log.jsonl (event: teammate_task_complete) to track team budget consumption.
When a soft limit is hit:
When a hard limit is hit:
default instead of heavy when a task doesn't require deep judgment/compact at milestones to reduce token waste/clear between unrelated tasks--- Budget Report ---
Delegations: 5 / 15
Heavy-tier: 2 / 8 (reviewer, security-reviewer)
Default-tier: 3 (implementer, researcher, doc-updater)
Fast-tier: 0
Estimated cost: ~$1.20
Recommendation: On track — no action needed