agentleFS
jaydubya818jaydubya818/MissionControl/skills/mission-control-request-approval/SKILL.md

mission-control-request-approval

Skill2 starsChanged 8 months agoMITView on GitHub

Requests human or dual-control approval for risky actions through the Convex api.approvals.request mutation. Use this skill when an action is YELLOW or RED risk tier — YELLOW needs a single approval, RED needs dual control (two decisions), and GREEN/LOW actions auto-approve. Poll status before acting.

Files
SKILL.md

Request Approvals

Risky actions must be approved before execution. Approvals are first-class records in Mission Control with an audit trail.

Risk Tiers

TierMeaningDecisions required
GREEN / LOWSafe, reversible0 — auto-approved on request
YELLOWNeeds sign-off1 approval
REDHigh blast radius2 approvals (dual control)

Request an Approval

Mutation: api.approvals.request
Args:
  requestorAgentId: Id<"agents">
  actionType: string          — Machine-readable action kind
  actionSummary: string       — One-line human-readable summary
  riskLevel: string           — "GREEN" | "LOW" | "YELLOW" | "RED"
  justification: string       — Why this action is needed
  projectId?: Id<"projects">
  taskId?: Id<"tasks">        — Related task (if any)
  toolCallId?: Id<"toolCalls">
  actionPayload?: any         — The exact payload you intend to execute
  estimatedCost?: number      — USD estimate
  rollbackPlan?: string       — How the action can be undone
  expiresInMinutes?: number   — Default 60
  idempotencyKey?: string     — Prevent duplicate requests on retries

Returns: { approval: Approval, created: boolean }

GREEN/LOW requests come back already APPROVED. YELLOW and RED come back PENDING; RED sets requiredDecisionCount: 2.

Check the Decision

Query: api.approvals.get
Args: { approvalId: Id<"approvals"> }

Query: api.approvals.listPending
Args: { projectId?: Id<"projects">, limit?: number }

The heartbeat response (mission-control-heartbeat) also includes pendingApprovals for your agent.

Approval status values: PENDING, ESCALATED, APPROVED, DENIED, EXPIRED, CANCELED.

Rules

  1. Never execute the action before the approval status is APPROVED. Denied or expired means do not proceed.
  2. Requests expire (default 60 minutes). Re-request with a new justification rather than acting on a stale approval.
  3. Move the related task to NEEDS_APPROVAL while waiting (mission-control-task-lifecycle).
  4. Provide rollbackPlan and estimatedCost for RED actions — reviewers need them to decide.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.