Building a GTM Engine with Agents

# stage-map.yaml: the order of GTM work and who approves what
stages:
  - id: research
    owner_agent: research-analyst
    inputs: [product_facts, watched_sources]
    outputs: [research_brief]          # every claim cited
    approval: human                    # you approve before anything reads it
    expires_after_days: 30

  - id: strategy
    needs_approved: [research]
    owner_agent: strategy-composer
    inputs: [research_brief, last_performance_brief, rejection_reasons]
    outputs: [positioning, narrative_arc, hook_bank]
    approval: human
    expires_after_days: 90

  - id: plan
    needs_approved: [strategy]
    owner_agent: plan-composer
    order: [icp, keyword_plan, channel_plan, calendar]
    outputs: [icp, keyword_plan, channel_plan, calendar]
    approval: human
    refresh: weekly

  - id: brief
    needs_approved: [plan]
    per: piece                          # one brief per planned piece
    approval: human                     # approve before generating anything

  - id: execute
    needs_approved: [brief]
    owner_agents: [content-writer, content-creator, publisher, outreach]
    approval: human_on_board
    first_n_held_per_channel: 5

  - id: measure
    owner_agent: performance-analyst
    outputs: [weekly_numbers_brief, lessons, proposed_rule_change]
    feeds: [plan]                       # back to strategy only when numbers say so
    approval: human

[
  { "id": "stage-gate",        "when": "before_tool",     "effect": "block",    "rule": "Upstream stage not approved" },
  { "id": "approved-only",     "when": "before_publish",  "effect": "block",    "rule": "Piece not approved by a human" },
  { "id": "claims-ledger",     "when": "before_publish",  "effect": "block",    "rule": "A number has no source in the claims ledger" },
  { "id": "taste-gate",        "when": "before_publish",  "effect": "block",    "rule": "Brand-voice score below 7.5" },
  { "id": "first-five-held",   "when": "before_publish",  "effect": "escalate", "rule": "Fewer than 5 approved sends on this channel" },
  { "id": "two-signals",       "when": "before_outreach", "effect": "block",    "rule": "Fewer than 2 independent buying signals" },
  { "id": "touch-budget",      "when": "before_outreach", "effect": "block",    "rule": "Contact already touched this week" },
  { "id": "suppression",       "when": "before_outreach", "effect": "block",    "rule": "Contact unsubscribed or bounced" },
  { "id": "spend-ceiling",     "when": "before_tool",     "effect": "escalate", "rule": "Daily spend ceiling reached or price unknown" },
  { "id": "bulk-action",       "when": "before_tool",     "effect": "warn",     "rule": "Action touches more than 50 records" }
]

CONTENT BRIEF
Piece:          [reel | image | carousel | article]   Date: [YYYY-MM-DD]
For:            [ICP segment, in one line]
Job to be done: [what the viewer wants at this moment]
One idea:       [the single point, in under 15 words]
Hook:           [first 2 seconds / first line]
Proof:          [claim + ledger source, or "no numbers"]
Shape:          [scene list / slide titles / H2 outline]
Call to action: [one action]
Channels:       [where it goes; same idea, adapted captions]
Must not:       [claims, names or tones to avoid]
Done means:     [what you'll check before approving]