GTMOps

Deterministic API tool layer for the GTM stack. 9 bash scripts that make your AI call APIs correctly the first time.

name: GTMOps
pack-id: themoatwriter-gtmops-v1.1.0
version: 1.1.0
author: themoatwriter
type: skill
platform: claude-code
keywords: gtm, instantly, slack, attio, pylon, gamma, bigin, n8n, serper, firecrawl, api, outbound, cold-email, crm, lead-generation

The Problem

Ask your AI assistant to generate a Gamma presentation. Here's what happens:

1. It greps your API key from .env manually
2. It writes a 30-line Python script with urllib.request
3. That fails (shell quoting issue). You hit Enter.
4. It rewrites the script. You hit Enter.
5. The script runs for 6 seconds, times out. You hit Enter.
6. It tries Authorization: Bearer (wrong, Gamma uses X-API-KEY). You hit Enter.
7. Cloudflare blocks with 403 because there's no User-Agent header. You hit Enter.
8. It adds the header but now the response field is wrong (id instead of generationId). You hit Enter.

Eight Enter presses. Maybe ten minutes. For something that should be:

gamma.sh generate --text "Your content here" --mode preserve

One Enter. Three seconds. Done.

This is the real cost of LLM API hallucination. It's not just wrong answers. It's death by Enter key, watching your AI assistant burn through attempts while you babysit every permission prompt.

We ran 8 common GTM operations without any tool layer. All 8 were silent failures: wrong field names, wrong auth patterns, fabricated endpoints, missing headers. The kind of failures where the API returns 200 OK but your data goes nowhere.

The Solution

A deterministic tool layer that sits between your LLM and the APIs:

9 bash wrapper scripts that bake in every gotcha from production use:

• Auth patterns (Bearer vs API key vs OAuth2 refresh)
• Correct field names (the ones that actually work, not the ones the LLM guesses)
• Error handling with actionable messages
• --dry-run to see the curl before executing
• --help with gotchas section on every tool

13 JSON payload templates that define exact shapes:

• Every template includes: JSON shape, curl command, tool command, field map, gotchas, example response

The difference: Your AI calls instantly.sh create-lead instead of building a curl command from scratch. One tool call, one Enter press, correct every time.

What You Use

9 bash scripts. One per API. --help, --dry-run, done.

Tool	API	What It Does
instantly.sh	Instantly v2	Lead creation, campaigns, SuperSearch (count/preview/enrich)
slack-connect.sh	Slack API	Channel creation, Slack Connect invites
attio.sh	Attio v2	Person/company upsert, record queries, field discovery
pylon.sh	Pylon	Account management, channel linking, issue tracking
gamma.sh	Gamma	Presentation/document generation with polling
bigin.sh	Zoho Bigin	Deal CRUD, contact management, OAuth2 auto-refresh
n8n.sh	n8n	Workflow management, execution debugging (VPS + Cloud)
serper.sh	Serper.dev	Google Search, Maps, Reviews, News, Images
firecrawl.sh	Firecrawl	Web scraping, search, site crawling, sitemap extraction

Also includes a guard hook (hooks/gtmops-guard.ts) that catches raw curl calls to covered APIs and reminds you to use the wrapper.

Installation

See INSTALL.md for the AI-assisted 6-phase installation wizard.

Quick start:

# 1. Clone
git clone https://github.com/themoatwriter/gtmops.git

# 2. Set up env
cp .env.example .env
# Fill in your API keys

# 3. Make tools executable
chmod +x src/tools/*.sh

# 4. Test
src/tools/instantly.sh --help
src/tools/attio.sh --help

What Makes This Different

Not just API docs. Every API has documentation. What they don't have:

1. Production-tested gotchas baked into code. The comment on line 22 of instantly.sh that says "Use campaign NOT campaign_id" exists because we spent 2 hours debugging a silent failure where leads went nowhere.

2. OnboardAPI.md for adding your own. A repeatable 6-phase workflow (Discover > Catalog Gotchas > Build Tool > Build Payloads > Register > Validate) so you can extend the stack with the same rigor.

3. One Enter, not fifteen. Your LLM calls a tested script instead of improvising curl commands. No permission prompt chains. No watching it fail and retry. It works the first time.

Test Results

8 common GTM operations, with and without GTMOps loaded:

#	Operation	Without	With	What Went Wrong
1	Create Instantly lead	FAIL	PASS	Used campaign_id instead of campaign. Lead went nowhere.
2	Instantly SuperSearch enrich	FAIL	PASS	Passed campaign ID instead of lead list ID
3	Upsert Attio person	FAIL	PASS	Used GET instead of PUT, wrong value format
4	Generate Gamma deck	FAIL	PASS	Used Bearer auth, missing User-Agent. 403 from Cloudflare.
5	Create Slack Connect invite	FAIL	PASS	Wrong scope, missing charset in Content-Type
6	Bigin deal stage update	FAIL	PASS	Wrong endpoint, missing {data: [...]} wrapper
7	Link Pylon channel	FAIL	PASS	Tried to create channel (Pylon can only link existing ones)
8	n8n workflow update	FAIL	PASS	Missing required settings field, got validation error

Without GTMOps: 0/8. Each failure meant 3-8 retry attempts. That's 30-50 Enter presses across the full run.

With GTMOps: 8/8. Eight tool calls. Eight Enter presses. Done.

Adding Your Own APIs

Follow the workflow in src/OnboardAPI.md:

1. Discover - Find docs, identify auth pattern, map key endpoints, test one call manually
2. Catalog Gotchas - Find the landmines before they blow up in production
3. Build Tool - Copy an existing script, modify for new API
4. Build Payloads - JSON templates with curl commands, field maps, gotchas
5. Register - Add to SKILL.md tables
6. Validate - --help, --dry-run, one real API call, valid JSON

What you extend: 13 payload templates

Payload templates document the exact JSON shapes, field maps, and gotchas for each API. They're reference docs for when you're onboarding a new API or need to understand what a tool does under the hood. You don't need to read them to use the tools.

Single-API operations: Lead create, campaign create, SuperSearch, subsequences, channel invites, record queries, DNC flows, email discovery, presentation generation, deal ops, contact create, workflow management, execution debugging, Google search/maps/reviews, web scraping/crawling, LLM chat completions.

Docs

• SKILL.md - Full tool/payload reference with trigger examples
• ApiReference.md - Consolidated endpoint docs and gotchas for all 9 APIs
• OnboardAPI.md - Standard workflow for adding new APIs to the stack

License

MIT