メニュー
Compliance and content checks that gate every outbound send. Used by the Composer (pre-draft) and the Cadence agent (pre-send).
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
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.
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.
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 | compliance |
| description | Compliance and content checks that gate every outbound send. Used by the Composer (pre-draft) and the Cadence agent (pre-send). |
Every outbound message — first-touch, reply, or templated cadence step — must pass these checks. The Composer runs them at draft time. The Cadence agent runs them again immediately before any auto-send (templates can have new variable values, so the check is per-send not per-template).
banned-phrases.md — exact strings the Composer must not emit.
Cliché outbound language plus voice-specific bans from
skills/outbound-voice/voice-dna.md.suppression-list.md — emails and domains never to contact.
Maintained as a flat list. Append on opt-out; never remove without an
explicit human decision.can-spam.md — CAN-SPAM requirements for US sends.gdpr.md — GDPR requirements for EU/UK sends.For every draft, in order:
suppression-list.md? If yes → hard fail. Move lead to
state: dormant with do_not_contact: true and never draft again.banned-phrases.md in the
draft body or subject? If yes → redraft. Maximum 3 redraft attempts;
after that, save to Drafts with hold_reason: voice and flag for
human review.OUTLOOK_FROMlawful_basis: in frontmatter:
legitimate-interest, consent, or contract). If absent → hard hold
until the operator records it.skills/kill-switches/triggers.md.hold_reason. The operator decides whether to send by hand.Every send must log to the prospect MD file's ## Audit section: