Skip to main content
Glama

Draft outreach

draft_outreach
Destructive

Produce two outreach draft variants (A/B) for a lead given an angle. Both drafts are warm and kind by design (P10) — variants differ in angle of helpfulness (subject hook, opening framing, call-to-action) not in tone. Drafts are written to lead_drafts as pending_review. Returns IDs + previews. Use after synthesize_lead_hypothesis to draft initial outreach.

Routing: CRM/sales → draft outreach copy for a lead (after synthesize_lead_hypothesis). Voice + reader-first + public audience labels load on this door.

[write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time. Call it on the first clear ask; the card is the yes — do not re-ask in chat.]

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
angleYesThe outreach angle to use (e.g., 'deeper_lp3_discovery', 'lighter_touch_different_hook', 'jurisdiction_clarification', 'kind_check_in'). Take from synthesize_lead_hypothesis.suggested_angle if unsure.
lead_idYesUUID of the lead.
reply_toNoOptional Reply-To address to carry on the eventual send (CONTRACT-1 agent thread address). Stamped into both drafts' metadata (best-effort — the metadata column is additive); send_lead_draft reads it at send time and passes it to send_email. Never changes what is drafted.
companyIdYesFreedomOS company id to act within (you must be a member). Required for company-scoped tools.
from_nameNoOptional sender display name (e.g., 'Acme Team', 'Alex at Acme'). Used in draft signature. When omitted, the CONTRACT-5 chain resolves it: the company's mcp_connections.resend.auth_config.from_name, else a generic 'Team'. Sequence callers pass the sequence's owning agent's name here (the top of the chain).
eligible_atNoOptional ISO timestamp — the earliest real time this draft may be sent (2026-07-13 send-timing gate). For a sequence step, pass now + that step's delay_hours (an ESTIMATE; the send-gate re-stamps it to the real value once the prior step actually resolves). Omitted → eligible immediately (the correct default for step 1 and for manual one-off drafts).
sequence_idNoOptional. The outreach_sequences.id the step belongs to. Pass it together with sequence_step_id to enable the A/B prior-stats bias — step ids repeat across sequences (step1…stepN), so stats are only comparable within one sequence. Also persisted on the draft row so the send-gate can resolve "the next step's draft" by an exact join instead of guessing. Omitted → no bias, no sequence linkage (manual one-off draft).
company_contextNoOptional short summary of the company the lead arrived at (e.g., 'Acme Health — pharmacy compounding compliance consulting'). Helps the model pitch correctly.
journey_summaryYesShort prose summary of what we know about this lead (their state, recent activity, what they engaged with). Used as context for the draft. Synthesis.intent_summary + 1-2 notes works well.
sequence_step_idNoOptional. If this draft is part of an auto-mode sequence step, pass the step_id from outreach_sequences. Otherwise omit (manual one-off draft).
regenerated_reasonNoOptional (regenerate-on-signal, 2026-07-15). When the sequencer re-drafts a not-yet-sent step after a meaningful lead signal (temperature flip to hot, a click), it passes a short human-readable reason (e.g. 'redrafted after they clicked'). Stamped into both drafts' metadata.regenerated_reason so the review card can show WHY the copy was refreshed. Never changes drafting logic — provenance only.
variant_b_guidanceNoOptional (CONTRACT-3). Sequence-designed seed for the B variant — a distilled subject+body angle persisted on the sequence step (steps jsonb, additive variant_b_guidance key). When present, variant_b is grounded in this guidance while variant_a stays the model's best independent take on the main angle. Omitted → both variants generated exactly as before.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed2 schema fields changed
    • changedInput schema / properties / company_context / description
      Previous value: -"Optional short summary of the company the lead arrived at (e.g., 'Conduit — pharmacy compounding consulting'). Helps the model pitch correctly."New value: +"Optional short summary of the company the lead arrived at (e.g., 'Acme Health — pharmacy compounding compliance consulting'). Helps the model pitch correctly."
    • changedInput schema / properties / from_name / description
      Previous value: -"Optional sender display name (e.g., 'Conduit Team', 'Lonnie at Conduit'). Used in draft signature. When omitted, the CONTRACT-5 chain resolves it: the company's mcp_connections.resend.auth_config.from_name, else a generic 'Team'. Sequence callers pass the sequence's owning agent's name here (the top of the chain)."New value: +"Optional sender display name (e.g., 'Acme Team', 'Alex at Acme'). Used in draft signature. When omitted, the CONTRACT-5 chain resolves it: the company's mcp_connections.resend.auth_config.from_name, else a generic 'Team'. Sequence callers pass the sequence's owning agent's name here (the top of the chain)."
  2. First observed

TDQS

A4.3/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Discloses the write behavior (drafts written as pending_review), the approval workflow ('first use may require a manager's approval...'), and design tone (warm/kind, differing only in angle). These go beyond the sparse annotations, which only mark destructiveHint and non-readOnly.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Core function is front-loaded in the first sentence, with return and sequencing immediately after. Some redundancy exists (routing repeats the synthesize_lead_hypothesis dependency) and internal jargon (P10, reader-first labels) adds noise, but it remains compact.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

No output schema exists, so the description rightly states returns ('IDs + previews') and side effects (pending_review). Given the 12-param schema that is fully described, plus approval and usage context, an agent has what it needs. Slight gap: no explicit mention that this tool does not send, but the name and pending_review state make that clear.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

All 12 parameters have schema descriptions (100% coverage), so the schema carries the semantic burden. The main description adds no parameter-specific details beyond referencing 'an angle', keeping this at the baseline.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb ('Produce'), resource ('two outreach draft variants (A/B) for a lead'), and input ('given an angle'). It also specifies the output sink ('written to lead_drafts as pending_review') and return ('IDs + previews'), clearly differentiating it from sending tools like send_lead_draft.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly says 'Use after synthesize_lead_hypothesis to draft initial outreach' and routes 'CRM/sales → draft outreach copy for a lead', giving clear context. It does not name exclusions or alternatives, so an agent must infer that sending is handled elsewhere.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources