Molly kickoff

Kick-off

Triggered when a recruiter @mentions Molly with a Notion JD page URL. Kick-off does two things: (1) build and write the Profile Spec, and (2) immediately run the first calibration cycle — find a batch of candidates (MOLLY_CALIBRATION_BATCH_SIZE, default 5) and post them to Slack for review.

Inputs

Extract the Notion JD page URL from the Slack message. The page ID is the last 32 hex characters of the URL.

Steps

1. Read the JD page

curl -s "https://api.notion.com/v1/blocks/JD_PAGE_ID/children?page_size=100" \
  -H "Authorization: Bearer $MOLLY_NOTION_API_KEY" \
  -H "Notion-Version: 2022-06-28"

Extract plain text from all blocks. The JD page contains: Job Description, Kick-off Notes, and Orienting Profiles (table). Read all three sections.

2. Build the Profile Spec

Synthesize the JD page content into a Profile Spec. The goal is to distill what makes a great candidate into something concrete enough to search for — not a restatement of the JD. The Exa Search Strategy section is especially important: it drives every candidate search, so make it specific.

Use this structure:

## Profile Spec: [Role Name]

### Ideal Candidate Summary
[1-2 sentences]

### Must-Haves
- ...

### Nice-to-Haves
- ...

### Evaluation Criteria
- [Criterion] ([weight]%): [what to look for]

### Target Companies
- ...

### Exa Search Strategy
[2-3 specific queries you will use; what signals indicate a strong match]

3. Create the Profile Spec page in Notion

Create a new page under the configured parent page ($MOLLY_NOTION_PARENT_PAGE_ID) titled "Profile Spec: [Role Name]" with the spec sections as content. Note the returned page ID — you'll need it for the tracker and button values.

curl -s -X POST "https://api.notion.com/v1/pages" \
  -H "Authorization: Bearer $MOLLY_NOTION_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Notion-Version: 2022-06-28" \
  --data @- <<EOF
{
  "parent": {"page_id": "$MOLLY_NOTION_PARENT_PAGE_ID"},
  "properties": {"title": {"title": [{"text": {"content": "Profile Spec: ROLE_NAME"}}]}},
  "children": [
    {"type": "heading_2", "heading_2": {"rich_text": [{"type": "text", "text": {"content": "Ideal Candidate Summary"}}]}},
    {"type": "paragraph", "paragraph": {"rich_text": [{"type": "text", "text": {"content": "CONTENT"}}]}},
    ... (repeat for each section)
  ]
}
EOF

4. Add a row to the Molly Tracker

curl -s -X POST "https://api.notion.com/v1/pages" \
  -H "Authorization: Bearer $MOLLY_NOTION_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Notion-Version: 2022-06-28" \
  --data @- <<EOF
{
  "parent": {"database_id": "$MOLLY_TRACKER_DB_ID"},
  "properties": {
    "Role name": {"title": [{"text": {"content": "ROLE_NAME"}}]},
    "JD Link": {"url": "JD_PAGE_URL"},
    "Profile Spec Link": {"url": "SPEC_PAGE_URL"},
    "Molly Status": {"status": {"name": "Calibrating"}},
    "Open date": {"date": {"start": "YYYY-MM-DD"}}
  }
}
EOF

5. Add Calibration Cycle 1 table to the Profile Spec page

Append a ## Calibration Cycle 1 heading and an empty table to the spec page. The table has 4 columns and starts with just the header row — candidate rows are added as feedback comes in.

Table structure:

{
  "object": "block",
  "type": "table",
  "table": {
    "table_width": 4,
    "has_column_header": true,
    "children": [
      {"object": "block", "type": "table_row", "table_row": {"cells": [
        [{"type": "text", "text": {"content": "Name"}}],
        [{"type": "text", "text": {"content": "Why it fits"}}],
        [{"type": "text", "text": {"content": "Team result"}}],
        [{"type": "text", "text": {"content": "Team feedback"}}]
      ]}}
    ]
  }
}

After appending, fetch the table block ID (you'll need it to add candidate rows later).

6. Search Exa for the calibration batch

Use the Exa Search Strategy from the spec. Run 2-3 queries with distinct angles. Search LinkedIn profiles only.

curl -s -X POST "https://api.exa.ai/search" \
  -H "Authorization: Bearer $MOLLY_EXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "YOUR_QUERY", "numResults": 10, "type": "neural", "includeDomains": ["linkedin.com"], "contents": {"text": {"maxCharacters": 500}}}'

Pick a full batch of non-duplicate matches (MOLLY_CALIBRATION_BATCH_SIZE, default 5). Exclude anyone below the minimum current-company tenure (MOLLY_MIN_TENURE_MONTHS, default 6 months) — global sourcing rule (see AGENTS.md); read the current role's start date / duration from the profile (fetch the full contents if the 500-char snippet doesn't reach it) and skip short-tenure candidates, over-fetching so you still land a full batch. For each survivor, record: name, current role + company, LinkedIn URL, 1-sentence fit rationale (cite specific evidence, not generic).

7. Add candidate rows to the Calibration Cycle 1 table

For each candidate in the batch, append a table row to the table block:

curl -s -X PATCH "https://api.notion.com/v1/blocks/TABLE_BLOCK_ID/children" \
  -H "Authorization: Bearer $MOLLY_NOTION_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Notion-Version: 2022-06-28" \
  -d '{"children": [{"object": "block", "type": "table_row", "table_row": {"cells": [
    [{"type": "text", "text": {"content": "NAME", "link": {"url": "LINKEDIN_URL"}}}],
    [{"type": "text", "text": {"content": "WHY_IT_FITS"}}],
    [{"type": "text", "text": {"content": ""}}],
    [{"type": "text", "text": {"content": ""}}]
  ]}}]}'

8. Post a single calibration cycle message to Slack

Post one message to the thread listing all candidates with a single "Review All Candidates" button. The button value encodes everything your Slack interactivity handler (see the README) needs to build the review modal: spec/JD page IDs, cycle number, role name, and the candidate list (name + LinkedIn URL for each).

Button value JSON (encode as string, keep compact — fits within Slack's 2000 char limit):

{
  "spec_page_id": "SPEC_PAGE_ID",
  "jd_page_id": "JD_PAGE_ID",
  "cycle_num": 1,
  "role_name": "ROLE_NAME",
  "candidates": [
    {"num": 1, "name": "Full Name", "url": "https://linkedin.com/in/..."},
    {"num": 2, "name": "Full Name", "url": "https://linkedin.com/in/..."},
    {"num": 3, "name": "Full Name", "url": "https://linkedin.com/in/..."},
    {"num": 4, "name": "Full Name", "url": "https://linkedin.com/in/..."},
    {"num": 5, "name": "Full Name", "url": "https://linkedin.com/in/..."}
  ]
}

Message Block Kit structure:

[
  {"type": "section", "text": {"type": "mrkdwn", "text": "📋 *ROLE_NAME* · <SPEC_URL|View Profile Spec>\n*Calibration Cycle N* — 5 candidates ready for review"}},
  {"type": "divider"},
  {"type": "section", "text": {"type": "mrkdwn", "text": "*1. <URL|Name>*\nRole at Company\n_Why they fit: RATIONALE_"}},
  {"type": "section", "text": {"type": "mrkdwn", "text": "*2. <URL|Name>*\nRole at Company\n_Why they fit: RATIONALE_"}},
  {"type": "section", "text": {"type": "mrkdwn", "text": "*3. <URL|Name>*\nRole at Company\n_Why they fit: RATIONALE_"}},
  {"type": "section", "text": {"type": "mrkdwn", "text": "*4. <URL|Name>*\nRole at Company\n_Why they fit: RATIONALE_"}},
  {"type": "section", "text": {"type": "mrkdwn", "text": "*5. <URL|Name>*\nRole at Company\n_Why they fit: RATIONALE_"}},
  {"type": "divider"},
  {"type": "actions", "elements": [
    {"type": "button", "text": {"type": "plain_text", "text": "Review All Candidates →"}, "style": "primary", "action_id": "open_review_modal", "value": "JSON_STRING"}
  ]},
  {"type": "context", "elements": [{"type": "mrkdwn", "text": "Click to approve/reject each candidate and add feedback. One submission fires Molly."}]}
]

curl -s -X POST "https://slack.com/api/chat.postMessage" \
  -H "Authorization: Bearer $MOLLY_SLACK_BOT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"channel": "CHANNEL", "thread_ts": "THREAD_TS", "blocks": [...]}'