메뉴
The canonical list of kill switches every agent checks at the top of every loop tick. If any trigger is active, the agent stops the relevant work.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
When to follow up, what to send, and when to stop. Defines pre-approved templated cadence steps that the Cadence agent can auto-fire under guardrails. Used by the Cadence agent.
Compliance and content checks that gate every outbound send. Used by the Composer (pre-draft) and the Cadence agent (pre-send).
How agents read from and write to Outlook via Microsoft Graph. Composer uses it to save Drafts. Cadence uses it to detect sends and replies. Read-only patterns work today via the M365 MCP; write paths are stubbed until a Graph write MCP or a small helper is wired.
How to research a prospect end-to-end — pull profile, pull company, surface recent signals, score ICP fit, write a dossier. Used by the Prospector agent.
How to write outbound messages in the Synozur firm voice. Used by the Composer agent. Anchored on a Voice DNA artifact extracted from real best-performing messages.
The markdown file schema for prospects. One file per lead at $PROSPECT_DIR/<slug>.md. Source of truth for state, dossier, drafts, conversation, and audit log. Read and written by all three agents.
| name | kill-switches |
| description | The canonical list of kill switches every agent checks at the top of every loop tick. If any trigger is active, the agent stops the relevant work. |
A kill switch is a condition that, when active, stops an agent from acting. They are not warnings. They are not soft signals. They are hard stops.
triggers.md — the exact triggers and their effects. Read at the top
of every agent loop and immediately before every send.Every agent — Prospector, Composer, Cadence — calls into this skill before doing any work, and again before any send. If any trigger is active, the agent reports it, logs to the prospect MD file (or to a global log), and exits cleanly.
compliance/Compliance checks gate content (banned phrases, suppression, CAN-SPAM). Kill switches gate behavior (campaign auto-pause, per-rep send caps, the global pause flag, error-rate thresholds). They look similar but they apply at different points in the flow: compliance is per-draft, kill switches are per-tick.
Keeping them separate makes it harder to accidentally remove a kill switch when editing compliance, and easier to add new ones without touching compliance.
A new trigger goes into triggers.md. Every agent's prompt already
imports this skill, so once the trigger is in the file, all three
agents check it. No code change required.