← Common RoomCONTENT HISTORY

Update to Common Room

Snapshot Nov 11, 2026 · 00:08 UTC · version 4.2.0

Collection source: downloaded plugin package. These snapshots do not have a confirmed matching collection source. Differences in file lists alone do not establish changes to the package.

WHAT CHANGED

Instructions updated for compose-outreach

Full description — Now says: signals, and enroll the contact in a sequence when the seller wants to send. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a...

Observed in instructions or declared skills. Runtime behavior has not been tested.

Save Common Room and check future changes in Following.

Up to 3 plugins free · no card required.

All fields and original values.

Full description

Before

Sep 30, 2026 · 22:52 UTC

signals. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]',

After

Nov 11, 2026 · 00:08 UTC

signals, and enroll the contact in a sequence when the seller wants to send. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', 'enroll [person] in [sequence]', 'send this', 'add them...

Skill instructions

Before

Sep 30, 2026 · 22:52 UTC

signals. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', or any outreach drafting request." Generate three personalized outreach formats — email, call script, and LinkedIn message ...

After

Nov 11, 2026 · 00:08 UTC

signals, and enroll the contact in a sequence when the seller wants to send. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', 'enroll [person] in [sequence]', 'send this', 'add them...

Supporting files

Before

Sep 30, 2026 · 22:52 UTC

[{"relative_path":"LICENSE","size_in_bytes":11358},{"relative_path":"NOTICE","size_in_bytes":173},{"relative_path":"agents/openai.yaml","size_in_bytes":442},{"relative_path":"references/my-company-context.md","size_in_bytes":1604},{"rela...

After

Nov 11, 2026 · 00:08 UTC

[{"relative_path":"LICENSE","size_in_bytes":11358},{"relative_path":"NOTICE","size_in_bytes":173},{"relative_path":"agents/openai.yaml","size_in_bytes":456},{"relative_path":"references/outreach-formats-guide.md","size_in_bytes":2745}]

Compare saved observations

Download comparison JSON

SKILL.md text changes

− removed · + added

--- before
+++ after
@@ -1,11 +1,11 @@
 ---
 name: compose-outreach
-description: "Generate personalized outreach messages using Common Room signals. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', or any outreach drafting request."
+description: "Generate personalized outreach messages using Common Room signals, and enroll the contact in a sequence when the seller wants to send. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', 'enroll [person] in [sequence]', 'send this', 'add them to a sequence', or any outreach drafting request."
 ---
 
 # Compose Outreach
 
-Generate three personalized outreach formats — email, call script, and LinkedIn message — grounded in Common Room signals for a specific company or contact.
+**Company context** means what the user's company sells and who it sells to (its ICP). Load it only when a step uses it (fit, ICP or positioning), not for plain lookups. `Me.company` gives the company's name, domains and a summary; if that's empty, look the company up as an Organization, then search the web; if neither helps, ask the user once per session.
 
 ## Data Source
 
@@ -15,58 +15,47 @@
 
 ### Step 1: Look Up the Target
 
-Use Common Room MCP tools to find and retrieve data for the target (company and/or specific contact). Pull:
-- Recent product activity and engagement signals
-- Community activity (posts, questions, reactions)
-- 3rd-party intent signals (job postings, news, funding)
-- Relationship history (prior contact, meetings, email opens)
+Retrieve the target company and/or contact from Common Room: recent product activity and engagement, community activity, third-party intent signals (job postings, news, funding) and relationship history (prior contact, meetings, email opens).
 
-If the user specified a person, run contact-level research. If only a company was given, identify the best contact to target based on title, engagement, and role.
+If the user named a person, research that contact. If only a company was given, pick the best contact by title, engagement and role.
 
 ### Step 2: Web Search for External Hooks (If CR Signals Are Thin)
 
-If CR returned strong signals (recent activity, engagement, product usage), those should drive personalization — skip web search. If CR signals are thin or the prospect has little CR activity, run a web search for external hooks:
-
-**What to search:**
+When Common Room returned strong signals, use them and skip web search. When they're thin, or the user asks for external hooks, search:
 - `"[company name]" funding OR acquisition OR launch OR announcement` — last 30 days
-- `"[contact full name]" "[company name]"` — look for recent articles, interviews, LinkedIn posts, or conference talks
-
-**Prioritize external hooks that are:**
-- Very recent (< 2 weeks) — the prospect is likely still thinking about it
-- Publicly visible — they know you could have seen it
-- Change-signaling — growth, new role, new product, new market
+- `"[contact full name]" "[company name]"` — recent articles, interviews, posts or talks
 
-If the user explicitly asks for web search or external hooks, run it regardless of CR signal richness.
+Prefer hooks under two weeks old that are public and signal change (growth, new role, new product, new market).
 
-### Step 3: Spark Enrichment (If Available)
+### Step 3: Spark (If Available)
 
-If Spark is available, run enrichment on the target contact to get persona classification, background, and influence signals. Use this to calibrate tone and message angle.
+Read the contact's Spark summary (`sparkSummary`) for persona, background and influence signals, and use it to set tone and angle.
 
 ### Step 4: Identify the Best Hooks
 
-From the signal data, identify the 1–3 strongest personalization hooks. Rank by:
-1. **Recency** — happened in the last 7–14 days
-2. **Specificity** — a concrete action they took, not a general trend
-3. **Relevance** — connects directly to a value your product delivers
+Pick the 1–3 strongest hooks, ranked by recency (last 7–14 days), specificity (a concrete action they took, such as a community question or a new hire wave, not "I noticed you're a customer" or an industry trend) and relevance to what the user's product delivers.
 
-Good hooks: posted a question in the community about X, just hired 5 engineers, recently started using [feature], company just raised Series B, trial nearing expiration, champion just changed jobs.
+### Step 5: Generate All Three Formats
 
-Bad hooks: "I noticed you're a customer" or generic industry trends.
+Write an email, a call script and a LinkedIn message from the strongest hooks, clearly labeled, following `references/outreach-formats-guide.md`. Warm and conversational for inbound or community signals; more formal for cold or executive outreach. With company context, ground the value bridge in the user's product and positioning.
 
-### Step 5: Generate All Three Formats
+### Step 6: Annotate Your Choices
 
-Use the strongest hooks to write all three formats. Each format has different constraints and conventions — follow the format-specific guidelines in `references/outreach-formats-guide.md`.
+After the drafts, add 2–4 sentences: which signals were used and why, any assumptions, and alternative angles.
 
-Always produce all three, clearly labeled.
+Then offer: "Want me to enroll [Name] in a sequence with this angle?"
 
-When the user's company context is available (see `references/my-company-context.md`), ground the value bridge and pitch in the user's specific product and positioning.
+### Step 7: Enroll (When the Seller Wants to Send)
 
-### Step 6: Annotate Your Choices
+Only when the seller asks to send or enroll. Never enroll unasked.
 
-After the three drafts, include a brief note (2–4 sentences) explaining:
-- Which signals were used and why they were chosen
-- Any assumptions made (e.g., inferred call objective)
-- Alternative angles if the primary hook doesn't land
+1. **Check the contact is saved.** Enrollment needs a Common Room contact ID. If the target came from web search or Prospector, offer to save them with commonroom_create_object first.
+2. **Pick the sequence.** List enrollment destinations with commonroom_list_action_destinations (action enroll_in_sequence). Recommend the one whose name or steps fit the hook from Step 4, and say why. Ask for the sending mailbox when the provider needs one; Zoom Engage sends as the seller by default. If the Sequences category is disabled, stop and say a workspace admin can allow it.
+3. **Reconcile the messages.** If the destination uses RoomieAI prompts, get each contact's draft with commonroom_get_message_drafts (up to 10 contacts per call). Compare each draft with the hook. Propose edits that work the hook in while keeping the admin's prompt and voice. Show the final snippets, and say that they fill RoomieAI slots in the sequence's own template (not the whole email), and that edits stay on the contact and are used in later enrollments too. If the destination has no RoomieAI prompts, say the sequence's existing content is sent, not the email drafted here.
+4. **Enroll.** One approval covers the contacts, sequence, sender and final snippets. Call commonroom_enroll_in_sequence with contactIds and the edited messages. Say that this sends real outreach on the sequence's schedule.
+5. **Check.** Read commonroom_list_action_logs with the runId and report each contact's outcome: succeeded, failed (with the reason) or pending. Don't retry automatically: a repeat starts a new run and can enroll contacts twice. Offer to rerun only the failed contacts, after the seller agrees.
+
+More than 10 contacts, or a filter instead of names: hand off to contact-actions.
 
 ## Output Format
 
@@ -104,35 +93,22 @@
 [2–4 sentences: which signals were used, why, and any alternative angles]
 ```
 
-## When Signal Data Is Sparse
-
-If Common Room returns minimal data on the target (e.g., just name, title, tags — no activity, no scores, no Spark):
-
-1. **Do not draft outreach from thin air.** Outreach grounded in fabricated signals is worse than no outreach.
-2. **Run web search first** — this becomes your primary personalization source. Look for recent news, LinkedIn posts, conference talks, company announcements.
-3. **If web search also returns little**, present what you have honestly and ask the user for context:
+After enrollment (Step 7), add:
 
 ```
-## Outreach for [Name / Company] — Limited Data
-
-**What I found:**
-[Only the real data from CR and web search]
+### Enrollment
+**Sequence:** [name] ([provider]) · **Sender:** [mailbox]
+**Why this sequence:** [one sentence tied to the hook]
 
-**I don't have enough signal to draft personalized outreach yet.** To write something strong, I'd need:
-- Recent activity or engagement signals
-- Context you have from prior conversations
-- A specific reason for reaching out now
+| Contact | RoomieAI field | Final message | Edited? |
+|---|---|---|---|
 
-Can you share any of the above?
+**Result:** [succeeded / failed with reason / pending] per contact
 ```
 
-## Quality Standards
+## When Signal Data Is Sparse
 
-- Every message must reference something specific — generic outreach is not acceptable output
-- Match tone to context: warm and conversational for inbound/community signals; more formal for cold/executive outreach
-- The LinkedIn message must be under 300 characters — no exceptions
-- The call script must be speakable naturally — read it aloud mentally to check rhythm
-- **Never fabricate signals** — only reference data retrieved from Common Room or web search
+If Common Room returns little on the target (no activity, no scores, no Spark), web search becomes the personalization source. If that also returns little, don't draft: show only what was found, say what's missing, and ask the user for prior-conversation context or a reason to reach out now. Never reference a signal that Common Room or web search didn't return, and don't offer enrollment without a real hook.
 
 ## Reference Files
 
Full technical diff · 3 changed fields

changed /description

BEFORE
"Generate personalized outreach messages using Common Room signals. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', or any outreach drafting request."
AFTER
"Generate personalized outreach messages using Common Room signals, and enroll the contact in a sequence when the seller wants to send. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', 'enroll [person] in [sequence]', 'send this', 'add them to a sequence', or any outreach drafting request."

changed /included_files

BEFORE
[
  {
    "relative_path": "LICENSE",
    "size_in_bytes": 11358
  },
  {
    "relative_path": "NOTICE",
    "size_in_bytes": 173
  },
  {
    "relative_path": "agents/openai.yaml",
    "size_in_bytes": 442
  },
  {
    "relative_path": "references/my-company-context.md",
    "size_in_bytes": 1604
  },
  {
    "relative_path": "references/outreach-formats-guide.md",
    "size_in_bytes": 2745
  }
]
AFTER
[
  {
    "relative_path": "LICENSE",
    "size_in_bytes": 11358
  },
  {
    "relative_path": "NOTICE",
    "size_in_bytes": 173
  },
  {
    "relative_path": "agents/openai.yaml",
    "size_in_bytes": 456
  },
  {
    "relative_path": "references/outreach-formats-guide.md",
    "size_in_bytes": 2745
  }
]

changed /skill_md_contents

BEFORE
"---\nname: compose-outreach\ndescription: \"Generate personalized outreach messages using Common Room signals. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', or any outreach drafting request.\"\n---\n\n# Compose Outreach\n\nGenerate three personalized outreach formats — email, call script, and LinkedIn message — grounded in Common Room signals for a specific company or contact.\n\n## Data Source\n\nUse the Common Room MCP server as the primary source. If its tools are unavailable or unauthenticated, tell the user that the Common Room connection is required and stop before claiming signal-based personalization. Ground every referenced signal in Common Room output or a cited web result.\n\n## Outreach Process\n\n### Step 1: Look Up the Target\n\nUse Common Room MCP tools to find and retrieve data for the target (company and/or specific contact). Pull:\n- Recent product activity and engagement signals\n- Community activity (posts, questions, reactions)\n- 3rd-party intent signals (job postings, news, funding)\n- Relationship history (prior contact, meetings, email opens)\n\nIf the user specified a person, run contact-level research. If only a company was given, identify the best contact to target based on title, engagement, and role.\n\n### Step 2: Web Search for External Hooks (If CR Signals Are Thin)\n\nIf CR returned strong signals (recent activity, engagement, product usage), those should drive personalization — skip web search. If CR signals are thin or the prospect has little CR activity, run a web search for external hooks:\n\n**What to search:**\n- `\"[company name]\" funding OR acquisition OR launch OR announcement` — last 30 days\n- `\"[contact full name]\" \"[company name]\"` — look for recent articles, interviews, LinkedIn posts, or conference talks\n\n**Prioritize external hooks that are:**\n- Very recent (< 2 weeks) — the prospect is likely still thinking about it\n- Publicly visible — they know you could have seen it\n- Change-signaling — growth, new role, new product, new market\n\nIf the user explicitly asks for web search or external hooks, run it regardless of CR signal richness.\n\n### Step 3: Spark Enrichment (If Available)\n\nIf Spark is available, run enrichment on the target contact to get persona classification, background, and influence signals. Use this to calibrate tone and message angle.\n\n### Step 4: Identify the Best Hooks\n\nFrom the signal data, identify the 1–3 strongest personalization hooks. Rank by:\n1. **Recency** — happened in the last 7–14 days\n2. **Specificity** — a concrete action they took, not a general trend\n3. **Relevance** — connects directly to a value your product delivers\n\nGood hooks: posted a question in the community about X, just hired 5 engineers, recently started using [feature], company just raised Series B, trial nearing expiration, champion just changed jobs.\n\nBad hooks: \"I noticed you're a customer\" or generic industry trends.\n\n### Step 5: Generate All Three Formats\n\nUse the strongest hooks to write all three formats. Each format has different constraints and conventions — follow the format-specific guidelines in `references/outreach-formats-guide.md`.\n\nAlways produce all three, clearly labeled.\n\nWhen the user's company context is available (see `references/my-company-context.md`), ground the value bridge and pitch in the user's specific product and positioning.\n\n### Step 6: Annotate Your Choices\n\nAfter the three drafts, include a brief note (2–4 sentences) explaining:\n- Which signals were used and why they were chosen\n- Any assumptions made (e.g., inferred call objective)\n- Alternative angles if the primary hook doesn't land\n\n## Output Format\n\n```\n## Outreach for [Name / Company]\n\n### 📧 Email\n\n**Subject:** [Subject line]\n\n[Email body — 3–5 sentences]\n\n---\n\n### 📞 Call Script\n\n**Opening:**\n[Opening line — conversational, 1–2 sentences]\n\n**Value Bridge:**\n[Why you're calling and why now — 2–3 sentences tied to a signal]\n\n**Ask:**\n[Single, low-friction ask — e.g., 15-minute call, specific question]\n\n---\n\n### 💼 LinkedIn Message\n\n[Under 300 characters. Warm, personal, no pitch.]\n\n---\n\n### Signal Notes\n[2–4 sentences: which signals were used, why, and any alternative angles]\n```\n\n## When Signal Data Is Sparse\n\nIf Common Room returns minimal data on the target (e.g., just name, title, tags — no activity, no scores, no Spark):\n\n1. **Do not draft outreach from thin air.** Outreach grounded in fabricated signals is worse than no outreach.\n2. **Run web search first** — this becomes your primary personalization source. Look for recent news, LinkedIn posts, conference talks, company announcements.\n3. **If web search also returns little**, present what you have honestly and ask the user for context:\n\n```\n## Outreach for [Name / Company] — Limited Data\n\n**What I found:**\n[Only the real data from CR and web search]\n\n**I don't have enough signal to draft personalized outreach yet.** To write something strong, I'd need:\n- Recent activity or engagement signals\n- Context you have from prior conversations\n- A specific reason for reaching out now\n\nCan you share any of the above?\n```\n\n## Quality Standards\n\n- Every message must reference something specific — generic outreach is not acceptable output\n- Match tone to context: warm and conversational for inbound/community signals; more formal for cold/executive outreach\n- The LinkedIn message must be under 300 characters — no exceptions\n- The call script must be speakable naturally — read it aloud mentally to check rhythm\n- **Never fabricate signals** — only reference data retrieved from Common Room or web search\n\n## Reference Files\n\n- **`references/outreach-formats-guide.md`** — detailed format rules, examples, and tone guidelines for each channel\n"
AFTER
"---\nname: compose-outreach\ndescription: \"Generate personalized outreach messages using Common Room signals, and enroll the contact in a sequence when the seller wants to send. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', 'enroll [person] in [sequence]', 'send this', 'add them to a sequence', or any outreach drafting request.\"\n---\n\n# Compose Outreach\n\n**Company context** means what the user's company sells and who it sells to (its ICP). Load it only when a step uses it (fit, ICP or positioning), not for plain lookups. `Me.company` gives the company's name, domains and a summary; if that's empty, look the company up as an Organization, then search the web; if neither helps, ask the user once per session.\n\n## Data Source\n\nUse the Common Room MCP server as the primary source. If its tools are unavailable or unauthenticated, tell the user that the Common Room connection is required and stop before claiming signal-based personalization. Ground every referenced signal in Common Room output or a cited web result.\n\n## Outreach Process\n\n### Step 1: Look Up the Target\n\nRetrieve the target company and/or contact from Common Room: recent product activity and engagement, community activity, third-party intent signals (job postings, news, funding) and relationship history (prior contact, meetings, email opens).\n\nIf the user named a person, research that contact. If only a company was given, pick the best contact by title, engagement and role.\n\n### Step 2: Web Search for External Hooks (If CR Signals Are Thin)\n\nWhen Common Room returned strong signals, use them and skip web search. When they're thin, or the user asks for external hooks, search:\n- `\"[company name]\" funding OR acquisition OR launch OR announcement` — last 30 days\n- `\"[contact full name]\" \"[company name]\"` — recent articles, interviews, posts or talks\n\nPrefer hooks under two weeks old that are public and signal change (growth, new role, new product, new market).\n\n### Step 3: Spark (If Available)\n\nRead the contact's Spark summary (`sparkSummary`) for persona, background and influence signals, and use it to set tone and angle.\n\n### Step 4: Identify the Best Hooks\n\nPick the 1–3 strongest hooks, ranked by recency (last 7–14 days), specificity (a concrete action they took, such as a community question or a new hire wave, not \"I noticed you're a customer\" or an industry trend) and relevance to what the user's product delivers.\n\n### Step 5: Generate All Three Formats\n\nWrite an email, a call script and a LinkedIn message from the strongest hooks, clearly labeled, following `references/outreach-formats-guide.md`. Warm and conversational for inbound or community signals; more formal for cold or executive outreach. With company context, ground the value bridge in the user's product and positioning.\n\n### Step 6: Annotate Your Choices\n\nAfter the drafts, add 2–4 sentences: which signals were used and why, any assumptions, and alternative angles.\n\nThen offer: \"Want me to enroll [Name] in a sequence with this angle?\"\n\n### Step 7: Enroll (When the Seller Wants to Send)\n\nOnly when the seller asks to send or enroll. Never enroll unasked.\n\n1. **Check the contact is saved.** Enrollment needs a Common Room contact ID. If the target came from web search or Prospector, offer to save them with commonroom_create_object first.\n2. **Pick the sequence.** List enrollment destinations with commonroom_list_action_destinations (action enroll_in_sequence). Recommend the one whose name or steps fit the hook from Step 4, and say why. Ask for the sending mailbox when the provider needs one; Zoom Engage sends as the seller by default. If the Sequences category is disabled, stop and say a workspace admin can allow it.\n3. **Reconcile the messages.** If the destination uses RoomieAI prompts, get each contact's draft with commonroom_get_message_drafts (up to 10 contacts per call). Compare each draft with the hook. Propose edits that work the hook in while keeping the admin's prompt and voice. Show the final snippets, and say that they fill RoomieAI slots in the sequence's own template (not the whole email), and that edits stay on the contact and are used in later enrollments too. If the destination has no RoomieAI prompts, say the sequence's existing content is sent, not the email drafted here.\n4. **Enroll.** One approval covers the contacts, sequence, sender and final snippets. Call commonroom_enroll_in_sequence with contactIds and the edited messages. Say that this sends real outreach on the sequence's schedule.\n5. **Check.** Read commonroom_list_action_logs with the runId and report each contact's outcome: succeeded, failed (with the reason) or pending. Don't retry automatically: a repeat starts a new run and can enroll contacts twice. Offer to rerun only the failed contacts, after the seller agrees.\n\nMore than 10 contacts, or a filter instead of names: hand off to contact-actions.\n\n## Output Format\n\n```\n## Outreach for [Name / Company]\n\n### 📧 Email\n\n**Subject:** [Subject line]\n\n[Email body — 3–5 sentences]\n\n---\n\n### 📞 Call Script\n\n**Opening:**\n[Opening line — conversational, 1–2 sentences]\n\n**Value Bridge:**\n[Why you're calling and why now — 2–3 sentences tied to a signal]\n\n**Ask:**\n[Single, low-friction ask — e.g., 15-minute call, specific question]\n\n---\n\n### 💼 LinkedIn Message\n\n[Under 300 characters. Warm, personal, no pitch.]\n\n---\n\n### Signal Notes\n[2–4 sentences: which signals were used, why, and any alternative angles]\n```\n\nAfter enrollment (Step 7), add:\n\n```\n### Enrollment\n**Sequence:** [name] ([provider]) · **Sender:** [mailbox]\n**Why this sequence:** [one sentence tied to the hook]\n\n| Contact | RoomieAI field | Final message | Edited? |\n|---|---|---|---|\n\n**Result:** [succeeded / failed with reason / pending] per contact\n```\n\n## When Signal Data Is Sparse\n\nIf Common Room returns little on the target (no activity, no scores, no Spark), web search becomes the personalization source. If that also returns little, don't draft: show only what was found, say what's missing, and ask the user for prior-conversation context or a reason to reach out now. Never reference a signal that Common Room or web search didn't return, and don't offer enrollment without a real hook.\n\n## Reference Files\n\n- **`references/outreach-formats-guide.md`** — detailed format rules, examples, and tone guidelines for each channel\n"
Full snapshot data
{
  "description": "Generate personalized outreach messages using Common Room signals, and enroll the contact in a sequence when the seller wants to send. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', 'enroll [person] in [sequence]', 'send this', 'add them to a sequence', or any outreach drafting request.",
  "included_files": [
    {
      "relative_path": "LICENSE",
      "size_in_bytes": 11358
    },
    {
      "relative_path": "NOTICE",
      "size_in_bytes": 173
    },
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 456
    },
    {
      "relative_path": "references/outreach-formats-guide.md",
      "size_in_bytes": 2745
    }
  ],
  "name": "compose-outreach",
  "skill_md_contents": "---\nname: compose-outreach\ndescription: \"Generate personalized outreach messages using Common Room signals, and enroll the contact in a sequence when the seller wants to send. Triggers on 'draft outreach to [person]', 'write an email to [name]', 'compose a message for [contact]', 'enroll [person] in [sequence]', 'send this', 'add them to a sequence', or any outreach drafting request.\"\n---\n\n# Compose Outreach\n\n**Company context** means what the user's company sells and who it sells to (its ICP). Load it only when a step uses it (fit, ICP or positioning), not for plain lookups. `Me.company` gives the company's name, domains and a summary; if that's empty, look the company up as an Organization, then search the web; if neither helps, ask the user once per session.\n\n## Data Source\n\nUse the Common Room MCP server as the primary source. If its tools are unavailable or unauthenticated, tell the user that the Common Room connection is required and stop before claiming signal-based personalization. Ground every referenced signal in Common Room output or a cited web result.\n\n## Outreach Process\n\n### Step 1: Look Up the Target\n\nRetrieve the target company and/or contact from Common Room: recent product activity and engagement, community activity, third-party intent signals (job postings, news, funding) and relationship history (prior contact, meetings, email opens).\n\nIf the user named a person, research that contact. If only a company was given, pick the best contact by title, engagement and role.\n\n### Step 2: Web Search for External Hooks (If CR Signals Are Thin)\n\nWhen Common Room returned strong signals, use them and skip web search. When they're thin, or the user asks for external hooks, search:\n- `\"[company name]\" funding OR acquisition OR launch OR announcement` — last 30 days\n- `\"[contact full name]\" \"[company name]\"` — recent articles, interviews, posts or talks\n\nPrefer hooks under two weeks old that are public and signal change (growth, new role, new product, new market).\n\n### Step 3: Spark (If Available)\n\nRead the contact's Spark summary (`sparkSummary`) for persona, background and influence signals, and use it to set tone and angle.\n\n### Step 4: Identify the Best Hooks\n\nPick the 1–3 strongest hooks, ranked by recency (last 7–14 days), specificity (a concrete action they took, such as a community question or a new hire wave, not \"I noticed you're a customer\" or an industry trend) and relevance to what the user's product delivers.\n\n### Step 5: Generate All Three Formats\n\nWrite an email, a call script and a LinkedIn message from the strongest hooks, clearly labeled, following `references/outreach-formats-guide.md`. Warm and conversational for inbound or community signals; more formal for cold or executive outreach. With company context, ground the value bridge in the user's product and positioning.\n\n### Step 6: Annotate Your Choices\n\nAfter the drafts, add 2–4 sentences: which signals were used and why, any assumptions, and alternative angles.\n\nThen offer: \"Want me to enroll [Name] in a sequence with this angle?\"\n\n### Step 7: Enroll (When the Seller Wants to Send)\n\nOnly when the seller asks to send or enroll. Never enroll unasked.\n\n1. **Check the contact is saved.** Enrollment needs a Common Room contact ID. If the target came from web search or Prospector, offer to save them with commonroom_create_object first.\n2. **Pick the sequence.** List enrollment destinations with commonroom_list_action_destinations (action enroll_in_sequence). Recommend the one whose name or steps fit the hook from Step 4, and say why. Ask for the sending mailbox when the provider needs one; Zoom Engage sends as the seller by default. If the Sequences category is disabled, stop and say a workspace admin can allow it.\n3. **Reconcile the messages.** If the destination uses RoomieAI prompts, get each contact's draft with commonroom_get_message_drafts (up to 10 contacts per call). Compare each draft with the hook. Propose edits that work the hook in while keeping the admin's prompt and voice. Show the final snippets, and say that they fill RoomieAI slots in the sequence's own template (not the whole email), and that edits stay on the contact and are used in later enrollments too. If the destination has no RoomieAI prompts, say the sequence's existing content is sent, not the email drafted here.\n4. **Enroll.** One approval covers the contacts, sequence, sender and final snippets. Call commonroom_enroll_in_sequence with contactIds and the edited messages. Say that this sends real outreach on the sequence's schedule.\n5. **Check.** Read commonroom_list_action_logs with the runId and report each contact's outcome: succeeded, failed (with the reason) or pending. Don't retry automatically: a repeat starts a new run and can enroll contacts twice. Offer to rerun only the failed contacts, after the seller agrees.\n\nMore than 10 contacts, or a filter instead of names: hand off to contact-actions.\n\n## Output Format\n\n```\n## Outreach for [Name / Company]\n\n### 📧 Email\n\n**Subject:** [Subject line]\n\n[Email body — 3–5 sentences]\n\n---\n\n### 📞 Call Script\n\n**Opening:**\n[Opening line — conversational, 1–2 sentences]\n\n**Value Bridge:**\n[Why you're calling and why now — 2–3 sentences tied to a signal]\n\n**Ask:**\n[Single, low-friction ask — e.g., 15-minute call, specific question]\n\n---\n\n### 💼 LinkedIn Message\n\n[Under 300 characters. Warm, personal, no pitch.]\n\n---\n\n### Signal Notes\n[2–4 sentences: which signals were used, why, and any alternative angles]\n```\n\nAfter enrollment (Step 7), add:\n\n```\n### Enrollment\n**Sequence:** [name] ([provider]) · **Sender:** [mailbox]\n**Why this sequence:** [one sentence tied to the hook]\n\n| Contact | RoomieAI field | Final message | Edited? |\n|---|---|---|---|\n\n**Result:** [succeeded / failed with reason / pending] per contact\n```\n\n## When Signal Data Is Sparse\n\nIf Common Room returns little on the target (no activity, no scores, no Spark), web search becomes the personalization source. If that also returns little, don't draft: show only what was found, say what's missing, and ask the user for prior-conversation context or a reason to reach out now. Never reference a signal that Common Room or web search didn't return, and don't offer enrollment without a real hook.\n\n## Reference Files\n\n- **`references/outreach-formats-guide.md`** — detailed format rules, examples, and tone guidelines for each channel\n"
}

SHA-256 of public snapshot: 00646b67aab8c741410a2467c81bea2ca2fb401188a6a96a3dbb3f8552b321ab