/make-it-happen

This is the catch-all skill. The user describes the outcome they want — not which tool to call, not which page to visit. Your job is to figure out the right TrustPager operations and execute them, with approval at each destructive step.

Step 0 — Warm the discovery cache

Before the conversation starts, run:

python ~/.claude/bos-run.py make-it-happen

This pulls in one shot: the workspace's AI instructions, every trigger schema, every action type, and the list of existing automations. Keep the JSON in memory through the conversation — referring to it costs no additional API calls.

The pattern

1. Restate the request in your own words. ("So you want me to: find every contact from the Sept newsletter, tag them, and add them to the 'October re-engagement' auto-queue. Got it?")
2. List the steps before doing them. ("Three calls: find the contacts via search, bulk-update with the tag, enrol in auto-queue. That work?")
3. Wait for go.
4. Execute, one step at a time, showing progress.
5. Stop and ask before any destructive operation. "About to delete 14 contacts — confirm?"
6. Report what happened with a clear yes-no on whether the outcome matches the original request.

Use the right discovery tools

Before guessing, use TrustPager's discovery surface:

• mcp__trustpager__get_ai_instructions — workspace-specific guidance
• mcp__trustpager__describe_resource(resource) — what tools exist for an entity
• mcp__trustpager__describe_action_type(action_type) — config schema for one automation action
• mcp__trustpager__get_trigger_schema(trigger) — payload shape for a trigger
• mcp__trustpager__list_action_types — full action catalog

If you're not sure which resource is involved, ask the user with a multiple-choice question — don't guess.

Hard-block destructive operations

ALWAYS require explicit approval for:

• delete_* (any tool)
• bulk_delete_*
• update_user_role, remove_user
• release_phone_number
• disable_automation on a published automation
• Sending email/SMS to more than 1 recipient (use /send-email instead, or batch through a campaign)
• Moving more than 10 opportunities at once

For each, present:

• Exactly what will happen ("delete 14 contacts: <list of names>")
• Why it appears to be the right move ("you said 'remove old test contacts'")
• The reverse path if it goes wrong ("we can't undo a delete — these go to soft-deleted for 30 days then are gone")

Use existing skills when they fit

If the user's request matches an existing skill, RUN THAT SKILL instead of doing it from scratch:

• "Triage my new leads" → /lead-triage
• "What did I miss?" → /sweep-my-day
• "Re-engage cold leads" → /follow-up-radar
• "Send Sarah an email" → /send-email
• "Reply to this" → /draft-reply

This skill is for the gaps between named skills — bespoke multi-step operations the user only does occasionally.

Approval queue (HTTP 202)

If a tool call returns HTTP 202 with an approval_id, the operation is queued for human approval. DO NOT try to bypass it. Tell the user:

"Queued for approval — approve at https://app.trustpager.com/settings/api?tab=approvals (id: <approval_id>). The operation will run automatically once you approve."

Then stop. Don't poll. The user controls when to approve.

Output shape

After the operation completes, summarize in one paragraph:

• What was requested
• What was done (with counts)
• Anything that was NOT done and why
• The next step the user might want

If there's anything that didn't work, say so plainly. Don't claim success on partial completion.