Agent activity log

A typed audit log of agent decisions, blockers, and outcomes — opt-in for users who are explicitly building an agent-driven workflow. The selling point: humans (and the next agent session) can read back what the agent did and why, without trusting a context-window summary that may be gone tomorrow. Pairs naturally with Airtable's role as a persistent agent substrate.

This skill owns the schema and disclosure language. Workflow skills (product-ops, sales-ops, marketing-ops, etc.) compose this skill rather than re-implementing the pattern; they trigger it when the user's intent surfaces an agent-driven workflow and pass through to this skill for the scaffolding + ongoing use.

When this fires

Trigger phrases — workflow skills should offer this to the user when any of these surface:

• "track what the agent is doing", "agent activity log," "audit log of agent decisions"
• "long-running workflow," "agent memory," "persistent state for the agent"
• "keep a record of what changed and why"
• Setup-mode invocations describing an agent-driven motion: "I want the agent to triage feedback every morning," "the agent should propose changes for me to approve," "set up a self-running workflow"

Surface proactively when the user's language signals they're building something the agent will run repeatedly or autonomously — not for one-shot interactions.

Disclosure (opt-in, not surveillance)