Claap Weekly Recap

Claap Weekly Recap

Reads every Claap recording from the last N days, extracts per-deal action items with verbatim quotes and Claap timestamp links, synthesizes weekly focus blocks, writes them to a Notion Kanban DB, and delivers a Slack summary. Designed to run unattended on a Sunday morning schedule after the user runs it once interactively to seed config.

You produce action items and focus blocks. You do NOT modify the underlying call recordings, edit existing Notion cards beyond the idempotency guard, or write to any database other than the GTM Action Items DB recorded in config.

When This Skill Applies

• "give me the week in calls"
• "what did prospects ask for this week"
• "draft action items from my Claap calls this week"
• "Sunday recap from sales calls"
• "set up a weekly recap from Claap into Notion + Slack"

NOT this skill:

• "summarize this one call" — use Claap's own per-call summary in the app
• "ask my call history a question right now" — use Claap MCP directly

Config File

The skill is driven by .claude/skills/claap-weekly-recap/.claap-config.json, created on first run by SETUP MODE. Secrets never live in this file. Only structural choices and non-sensitive IDs:

{
  "version": 1,
  "claap_tool_prefix": "mcp__claap__",
  "gtm_action_items_data_source_id": "collection://<YOUR_DATA_SOURCE_UUID>",
  "notion_db_url": "https://www.notion.so/<workspace>/<db-id>",
  "slack_delivery": {
    "mode": "mcp_user",
    "target": "<U_YOUR_SLACK_ID>"
  },
  "lookback_days": 7,
  "min_call_minutes": 3,
  "extraction_signals": [
    "objection",
    "competitor_mention",
    "feature_request",
    "action_item",
    "next_step_promised"
  ],
  "kanban_statuses": ["Backlog", "Focus This Week", "Done", "Archived"],
  "extra_properties": [],
  "prior_recap_history_path": "~/.gtm-os/state/claap-recap-history.md",
  "model": "claude-sonnet-4-6",
  "max_turns": 100
}

Allowed claap_tool_prefix values:

• mcp__claap__ (self-hosted / project-scoped Claap MCP)
• mcp__claude_ai_Claap__ (claude.ai connector)

Allowed slack_delivery.mode values:

• mcp_user — target is a Slack user ID like <U_YOUR_SLACK_ID>
• mcp_channel — target is a channel ID like <C_YOUR_CHANNEL_ID>
• webhook — target is the literal string env:SLACK_WEBHOOK_URL. The webhook URL itself is never written to config; only the env-var name.

Secrets (must live in env or shell rc, never in config.json):

• CLAAP_API_KEY — required at MCP launch time
• SLACK_WEBHOOK_URL — only required if slack_delivery.mode == "webhook"

Process

SETUP MODE (interactive)

Enter SETUP MODE if either of these is true:

• The config file .claude/skills/claap-weekly-recap/.claap-config.json does not exist, OR
• A required runtime input (e.g. gtm_action_items_data_source_id) cannot be resolved AND no config file exists.

If config exists but is incomplete (e.g. version bump added a field), prompt only for the missing pieces — don't restart from step 1.

Walk the user through these steps one question at a time. Wait for an answer before moving on. Confirm each persisted choice.

1. Resolve the Claap MCP tool prefix.

Call ListMcpResourcesTool. Look for tool names starting with either:

• mcp__claap__ (e.g. mcp__claap__list_workspaces)
• mcp__claude_ai_Claap__ (e.g. mcp__claude_ai_Claap__list_workspaces)

If exactly one is present, record it as claap_tool_prefix and confirm with the user. If both are present, ask which to use. If neither is present, print the .mcp.json snippet below and offer to write it to <workspace>/.mcp.json (creating or merging into mcpServers):

{
  "mcpServers": {
    "claap": {
      "url": "https://api.claap.io/mcp",
      "type": "http",
      "headers": {
        "Authorization": "Bearer ${CLAAP_API_KEY}"
      }
    }
  }
}

If the user declines, halt SETUP MODE with: "Add the Claap MCP to your config and restart Claude Code, then re-run me." If the user accepts, write the snippet, then halt with: "Restart Claude Code so the MCP loads, then re-run me — I'll pick up where we left off."

2. Confirm CLAAP_API_KEY is in env.

Run printenv CLAAP_API_KEY via Bash and read the exit code only (do not echo the value into chat under any circumstance). If it's set, say "CLAAP_API_KEY detected — moving on." and continue.

If it's missing, ask the user to paste their key. Read it into a local variable but never repeat it back in chat. Offer two persistence paths:

• Append export CLAAP_API_KEY=<masked> to ~/.zshenv (zsh) or ~/.bashrc (bash) — pick based on $SHELL.
• Skip persistence (key only available in current session).

Confirm the choice before writing. If user declines persistence, warn that scheduled runs will fail unless the key is in env at MCP launch. Never put the key in .claap-config.json. Never echo the key after capture.

3. Detect the Notion MCP.

Look for mcp__claude_ai_Notion__notion-fetch or mcp__notion__notion-fetch. If neither is loaded, halt with:

"Notion MCP isn't loaded. Connect the claude.ai Notion connector (Settings → Connectors) or add mcp.notion.com to your .mcp.json, then restart Claude Code and re-run me."

Record which prefix worked under notion_tool_prefix (the skill assumes the same notion-* tool names suffix on both).

4. Ask for the Notion parent page.

"Paste the Notion URL or UUID of the page where the GTM Action Items DB should live."

Resolve via notion-fetch to validate it exists and is writable. If resolution fails, ask again. Persist nothing yet — only the resolved page_id is used in step 8.

5. Ask which signals to extract.

"Which signals should I extract from transcripts? Press Enter for the default 5, or paste a comma-separated list."

Default: objection, competitor_mention, feature_request, action_item, next_step_promised.

Persist under extraction_signals.

6. Ask for Kanban column names.

"Which Kanban columns do you want? Press Enter for the default (Backlog, Focus This Week, Done, Archived) or paste a comma-separated list."

If the user supplies custom columns, ask for a color per column (must be one of: default, gray, brown, orange, yellow, green, blue, purple, pink, red). Persist column names under kanban_statuses.

7. Ask for extra properties (optional).

"Any extra properties on top of the default 10? Press Enter to skip, or paste lines like Owner: rich_text / Priority: select(High:red, Med:yellow, Low:gray)."

Persist under extra_properties as an array of {name, type, options?}.

8. Create the database.

Build a notion-create-database body following the contract in references/notion-db-schema.md. Parameterize:

• parent.page_id ← the resolved page from step 4
• Status options ← kanban_statuses (color per user input, default mapping when defaults are used)
• Add any extra_properties to the initial_data_source.properties map

Do not use pseudo-SQL. The JSON body in references/notion-db-schema.md is the only valid input shape.

On success, capture:

• The DB URL → notion_db_url
• The data_source UUID → gtm_action_items_data_source_id (store as collection://<uuid> for stability)

9. Discover property IDs and create the Kanban view.

Call notion-fetch on the new data source. From the response, extract property_id for Status, Title, Type, Deal / Lead, Verbatim Quote, Claap Link, Source Call Date, Week, Surfaced By, Due.

Call notion-create-view with the board configuration in references/notion-db-schema.md, grouped by Status, showing the properties above.

10. Slack delivery.

"How should I deliver the weekly Slack summary?"

Offer three modes:

a. DM to a Slack user — ask for the user ID. If the user provides a handle or name instead, resolve via slack_search_users (record the resolved U... ID). Persist as slack_delivery: { mode: "mcp_user", target: "<U_YOUR_SLACK_ID>" }.

b. Post to a Slack channel — ask for the channel ID (or resolve from name via slack_search_channels). Persist as slack_delivery: { mode: "mcp_channel", target: "<C_YOUR_CHANNEL_ID>" }.

c. Incoming webhook — instruct the user to create an incoming webhook in Slack and export it as an env var:

export SLACK_WEBHOOK_URL="https://hooks.slack.com/services/..."

Offer to append export SLACK_WEBHOOK_URL=<masked> to the user's ~/.zshenv (zsh) or ~/.bashrc (bash). Confirm before writing. Never echo the URL back in chat after capture. Persist as slack_delivery: { mode: "webhook", target: "env:SLACK_WEBHOOK_URL" }. The literal URL is never written to config.

11. Write .claap-config.json.

Persist all collected non-secret fields to .claude/skills/claap-weekly-recap/.claap-config.json (the skill folder's .gitignore already excludes this file).

Final shape:

{
  "version": 1,
  "claap_tool_prefix": "<mcp__claap__ or mcp__claude_ai_Claap__>",
  "notion_tool_prefix": "<mcp__claude_ai_Notion__ or mcp__notion__>",
  "gtm_action_items_data_source_id": "collection://<YOUR_DATA_SOURCE_UUID>",
  "notion_db_url": "https://www.notion.so/<workspace>/<db-id>",
  "slack_delivery": { "mode": "...", "target": "..." },
  "lookback_days": 7,
  "min_call_minutes": 3,
  "extraction_signals": ["objection", "competitor_mention", "feature_request", "action_item", "next_step_promised"],
  "kanban_statuses": ["Backlog", "Focus This Week", "Done", "Archived"],
  "extra_properties": [],
  "prior_recap_history_path": "~/.gtm-os/state/claap-recap-history.md",
  "model": "claude-sonnet-4-6",
  "max_turns": 100
}

Print: "Setup complete. Run me again to generate this week's recap."

Do not proceed to the weekly run in the same session. The schedule infra needs the persisted config; running once and exiting cleanly catches mis- configurations before they hit production.

NORMAL MODE (weekly run — config present)

Load .claap-config.json. Substitute {{claap_tool_prefix}} with the config value in every reference below before calling the tool. (E.g. for {{claap_tool_prefix}}list_workspaces, call mcp__claap__list_workspaces or mcp__claude_ai_Claap__list_workspaces.)

Inputs

Tools Used

Claap MCP (read). Claap exposes 7 tools under the claap_tool_prefix. The four relevant ones:

{{claap_tool_prefix}}search_companies, search_contacts, search_deals are available if you need to resolve participants → deals when the recording doesn't carry a dealId.

Notion (write).

• notion-fetch on the target DS to confirm the schema and discover the current select option set.
• notion-fetch (with filter) to check for existing pages this week before creating new ones (idempotency guard — Step 6).
• notion-create-pages to add cards (batch ≤40 per call).
• notion-update-data-source to seed missing select options (always include { "id": "<existing-id>" } entries for options you want to keep — see references/notion-db-schema.md).

Slack (notify).

Pick the path that matches slack_delivery.mode in config:

• mcp_user → slack_send_message with channel_id = slack_delivery.target (the user ID; Slack treats DMs that way).

• mcp_channel → slack_send_message with channel_id = slack_delivery.target.

• webhook → resolve env var named in target (e.g. SLACK_WEBHOOK_URL) via Bash. If the env var is empty, hard-stop with a clear error. Then POST the message body using the temp-file pattern (never use process substitution <(...) — fragile in cron / launchd / claude -p shells):

  printf '{"text": %s}' "$(jq -Rs . < /tmp/recap-body.txt)" > /tmp/recap.json
  curl -X POST -H 'Content-Type: application/json' \
    --data-binary @/tmp/recap.json "$SLACK_WEBHOOK_URL"

If the configured delivery mode can't be resolved (env var missing, MCP not loaded), hard-stop. Do not silently skip the notification.