Update to ACA

{
  "description": "Design one controlled ACA experiment for outbound or content. Use when the user wants to improve reply rate, test an angle, split audiences, A/B test copy, compare channels, or plan experiments.",
  "included_files": [],
  "name": "aca-experiment-design",
  "skill_md_contents": "---\nname: aca-experiment-design\ndescription: \"Design one controlled ACA experiment for outbound or content. Use when the user wants to improve reply rate, test an angle, split audiences, A/B test copy, compare channels, or plan experiments.\"\nlicense: MIT\ncompatibility: Requires the ACA connector (OAuth sign-in, no API key).\nmetadata:\n  author: ACA\n  version: \"1.1\"\n  homepage: https://www.automatedclientacquisition.com/mcp\n---\n\n# ACA experiment design\n\nDesign a clean experiment with one variable, one metric, and a stopping rule, grounded in the workspace's real baseline.\n\n## Rules\n\n- `add_contacts_to_list` can trigger the target list's existing enrichment, campaign enrollment, external outreach, and configured webhooks. Before adding contacts, review that list's automation settings at `/leads/lists` and confirm the intended contacts and side effects. If the connector cannot expose those settings, ask the user to verify them in ACA; do not assume a list write is draft-only.\n- Use the ACA connector only. Never ask for vendor API keys or run SQL/API calls.\n- Change one variable at a time.\n- Prefer small tests before scaling.\n- Write the experiment plan before anything changes.\n- Baselines come from stored counts as returned. `get_campaign_metrics` returns counts, never rates. Only compute a rate when the matching denominator came back too, and show both numbers. If there is no usable baseline, say so and make the first run the baseline.\n- Never invent results.\n- Ask before any write. Editing sequences, changing campaign settings, enrolling, and activating happen in the ACA app.\n\n## Workflow\n\n### 1. Read the current state\n\nCall `get_started`, then:\n\n- `list_campaigns` and `get_campaign_metrics` for the campaigns in scope\n- `list_sequences` and `get_sequence` for email tests\n- `search_conversations` with `has_inbound_message: true` for reply volume context\n- `list_lead_lists` for audience size\n- `list_content_generation_jobs` for content tests\n\n### 2. Pick the experiment type\n\n- Audience\n- Angle\n- CTA\n- Channel\n- Sequence timing\n- Content topic\n\n### 3. Define the test\n\n- Hypothesis\n- Control (what runs now, with its stored baseline counts)\n- Variant (the one thing that changes)\n- Metric (a count with its denominator, for example \"replies of contacts messaged\")\n- Sample size per arm (a list size you can actually fill, from `list_lead_lists`)\n- Decision rule (the difference that counts as a win, and what happens if it is a tie)\n- Review date\n\n### 4. Write the plan\n\nShow it in chat. Offer to let the user save it at `/assets?tab=strategies`.\n\n### 5. Route to the executor (after approval)\n\n- Copy variable: `aca-copy-variants` (copy written in chat; built at `/email/sequences` or `/campaigns/{id}`)\n- Audience variable: `aca-find-leads` or `aca-lead-quality` (variant list via `create_lead_list` and `add_contacts_to_list`)\n- Channel or new LinkedIn arm: `aca-launch-outreach` (inactive draft via `create_linkedin_campaign_draft`)\n- Content variable: `aca-content-week`\n\n## Output format\n\n```text\nExperiment: {name}\nHypothesis: {hypothesis}\nControl: {control} (baseline: {stored counts})\nVariant: {variant}\nMetric: {count} of {denominator}\nSample size: {n} per arm\nDecision rule: {rule}\nReview date: {date}\nNext action: {skill} → {app link if the change is app-only}\n```\n\n## Skill chaining\n\nPreserve the workspace, relevant IDs, the user's brief, and approval state when continuing into another ACA skill. If the user asked for execution and a downstream condition is met, continue into the next skill; otherwise end with the handoff block.\n\n**Upstream**\n- Called by `aca-weekly-rhythm`, `aca-positive-reply-scoring`, `aca-copy-variants`, or optimization requests.\n\n**Auto-continue conditions**\n- Copy variable: continue to `aca-copy-variants`.\n- Audience variable: continue to `aca-find-leads` or `aca-lead-quality`.\n- Content variable: continue to `aca-content-week`.\n- Launch-ready test: continue to `aca-launch-outreach`.\n\n**Stop before chaining when**\n- Applying experiment changes or creating records the user has not approved.\n- The change happens in the ACA app (sequence edits, enrollment, activation).\n\n**Downstream skills**\n- `aca-copy-variants`: create message variants.\n- `aca-find-leads`: build the audience variant.\n- `aca-lead-quality`: segment the audience variant.\n- `aca-content-week`: create the content variant.\n- `aca-launch-outreach`: launch the approved experiment.\n\n**Handoff block**\n\n```text\nChain state: {continue|needs_approval|blocked|complete}\nNext skill: {aca-skill-name|none}\nReason: {why this handoff is or is not needed}\nCarry forward: {workspace, campaign_id, sequence_id, lead_list_id, baseline counts, experiment plan, approvals, constraints}\n```\n\n## ACA tools used\n\n- `get_started`\n- `list_campaigns`, `get_campaign_metrics`\n- `list_sequences`, `get_sequence`\n- `search_conversations`\n- `list_lead_lists`\n- `list_content_generation_jobs`\n\n## ACA app pages\n\n- Sequences: `https://www.automatedclientacquisition.com/email/sequences`\n- Campaigns: `https://www.automatedclientacquisition.com/campaigns/{id}`\n- Save the plan: `https://www.automatedclientacquisition.com/assets?tab=strategies`\n"
}