Install
$ agentstack add skill-hainrixz-claude-ads-ads-start ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Guided onboarding wizard
You are running the first-run / re-entry wizard that produces and maintains ~/.claude-ads/profile.json. This is the canonical entry point for new users and the only place the user re-edits their saved context. The wizard never auto-invokes other /ads commands — at the end it suggests the next command and exits.
Core rules
- One question at a time. Each step uses ONE
AskUserQuestionblock with
2–4 options, then writes the answer to the profile via python3 scripts/profile.py set BEFORE asking the next question. Partial completion must be recoverable.
- Step-by-step verification. For OAuth / MCP setup, dictate one step at a
time, wait for the user's "done", run a verification call, diagnose errors. Cap retries at 3 per platform before offering "skip + come back later".
- Never store secrets in the profile. Tokens, API keys, refresh tokens
live in environment variables only. The profile records api_key_present: true/false, never the value.
- The wizard is inert at the end. Suggest the next command in plain text,
do NOT execute it.
Subcommand routing
Read the user's argument (if any):
| Arg | Behavior | |-----|----------| | (none) | Phase A — detect first-run vs re-entry. | | status | Run python3 scripts/profile.py get, print a 5-line summary, exit. | | edit | Skip Phase A welcome, jump to the edit selector in Phase A.4. | | reset | Confirm with AskUserQuestion → call python3 scripts/profile.py reset --yes → restart from Phase A. |
Phase A — Detect first-run vs re-entry
A.1 Run python3 scripts/profile.py get and capture exit code.
- Exit
2(no profile) → first-run: print the welcome banner below, then go
to Phase B.
- Exit
0(profile exists) → re-entry: print a 5-line summary (industry,
spend bracket, active platforms, last command) and go to A.4.
A.2 First-run welcome (print verbatim, adapt to user's language):
Welcome to claude-ads. I'll walk you through 4 quick steps:
1. Tell me about your business (~1 min)
2. Connect Meta / Google / TikTok — one at a time, step by step (~2-10 min each)
3. Optional: Zernio for publishing creatives, Meta for Developers
4. Pick your first action
Nothing leaves your machine. Tokens go to env vars, not the profile.
You can re-run /ads start edit anytime to change settings, or /ads start reset
to start over.
A.3 First-run: run python3 scripts/profile.py init and continue to Phase B.
A.4 Re-entry edit selector — AskUserQuestion:
> "What do you want to change?"
Business context— re-run Phase B.Meta connection— re-run Phase C for Meta only.Google connection— re-run Phase C for Google only.TikTok connection— re-run Phase C for TikTok only.Zernio publishing— re-run Phase D Zernio block.
(AskUserQuestion will add "Other" — treat it as cancel.)
After completing the selected edit, write last_command_at via python3 scripts/profile.py set last_command_at $(date -u +%FT%TZ) and exit with a one-line "Saved." confirmation.
Phase B — Business context (first-run or edit context)
Four sequential AskUserQuestion blocks. Save EACH answer immediately.
B.1 Industry — AskUserQuestion (single-select)
Options: Ecommerce · Local service · Real estate · Healthcare · Finance · Agency / multiple clients · Other.
Map to profile values: ecommerce, local-service, real-estate, healthcare, finance, agency, other.
Then: python3 scripts/profile.py set context.industry .
B.2 Monthly ad spend — AskUserQuestion (single-select)
Options: Under $500/mo · $500–$2,000/mo · $2,000–$10,000/mo · $10,000+/mo. (User can pick "Other" and type an exact figure — coerce to a number.)
Map bracket → midpoint USD: 250, 1250, 6000, 25000. If user gave an exact number, use it.
Then: python3 scripts/profile.py set context.monthly_spend_usd .
B.3 Primary goal — AskUserQuestion (single-select)
Options: Sales / revenue · Leads / signups · App installs · Phone calls · Brand awareness.
Map → sales, leads, app-installs, calls, brand.
Then: python3 scripts/profile.py set context.primary_goal .
B.4 Active platforms — AskUserQuestion (multiSelect: true)
> "Which platforms do you actively run (or plan to run) ads on?"
Options: Meta (Facebook + Instagram) · Google Ads (incl. YouTube) · TikTok Ads.
Require ≥1. Map to ["meta","google","tiktok"] subset.
Then: python3 scripts/profile.py set context.active_platforms ''.
Phase C — Per-platform connection wizard
For each platform in context.active_platforms (in order Meta → Google → TikTok, only those selected in B.4), run the loop below.
C.1 Tier selection — AskUserQuestion (single-select)
> "How do you want to connect ``?"
| Option | Save as connections..tier | Notes | |--------|---------------------------------|-------| | MCP server (recommended) | mcp_claude_ai for Meta · mcp_cohnen for Google · mcp_adsmcp for TikTok | Capa 1. Quickest. | | Direct API (free, OAuth once) | direct_api | Capa 2. Full control. | | Manual exports for now | manual | Capa 3. Always works, no setup. | | Skip — set this up later | skipped | Leave unconfigured. |
If manual or skipped: write the tier, write connected_at to now, skip C.2-C.4, continue to the next platform.
C.2 Load the platform setup walkthrough
Read the relevant reference file:
| Platform | Reference | |----------|-----------| | Meta | ads/references/setup-meta.md | | Google | ads/references/setup-google.md | | TikTok | ads/references/setup-tiktok.md |
Each file has Sections A (MCP), B (Direct API), C (Manual). Jump to the section that matches the user's tier choice. Within that section, walk the steps in order.
C.3 Step-by-step verified walkthrough
For each STEP in the chosen section:
- Print the step verbatim: its Goal, the exact URL/action under "Do
this", and what they should "Expect to see".
AskUserQuestion→ "Done with step N?" with options:
Done — next stepHit an errorCancel and skip this platform
- On
Done: continue to the next step. - On
Hit an error: read the step's "On error" branch and ask the user to
describe what they saw. Suggest the most likely fix from the reference. Retry the step. Cap at 3 attempts → offer to skip.
- On
Cancel: setconnections..tier = "skipped"and continue to next
platform.
After the LAST step, run the verification call:
| Platform | Tier A (MCP) verification | Tier B (Direct API) verification | |----------|---------------------------|----------------------------------| | Meta | Call mcp__claude_ai_Facebook__ads_get_ad_accounts | python3 scripts/api/meta_fetch.py --account-id --what account | | Google | Call mcp__cohnen_google_ads__list_customers (if available) — else ask user to paste customer_id | python3 scripts/api/google_fetch.py --customer-id --what account | | TikTok | List advertiser via MCP tool — else ask user to paste advertiser_id | python3 scripts/api/tiktok_fetch.py --advertiser-id --what account |
(If the MCP tool isn't present in this session — e.g., the user hasn't enabled it on claude.ai — fall back to asking the user to paste the account ID manually and skip the live call. Record last_verified_at: null so /ads next knows to re-verify next time.)
C.4 Save the connection
On verification success:
python3 scripts/profile.py set connections..ad_account_id ""
python3 scripts/profile.py set connections..connected_at "$(date -u +%FT%TZ)"
python3 scripts/profile.py set connections..last_verified_at "$(date -u +%FT%TZ)"
Print ✓ confirmation with the account ID. Continue to next platform.
Phase D — Optional account-creation walkthroughs
D.1 Meta for Developers (only if Meta tier == direct_api)
AskUserQuestion:
> "Do you already have a Meta for Developers account? (Required for the > direct Meta API path.)"
Options: Yes — I have one · No — walk me through it · Skip.
On No: load ads/references/setup-meta-developers.md and run the step-by-step loop (same pattern as C.3). When the user has created their app and added the Marketing API product, return to C.3 step 1 of the Direct API section in setup-meta.md.
D.2 Zernio publishing (asked once at the end of Phase C, regardless of tier)
AskUserQuestion:
> "Do you want to enable /ads publish for posting generated creatives to 14+ > social networks? Zernio's free tier covers 2 connected accounts forever (no > credit card)."
Options: Yes — walk me through Zernio signup · Maybe later — skip · I already have an account.
Yes: loadads/references/setup-zernio.md, run step loop. On success
write zernio.signed_up = true. If user reports they saved the API key to ZERNIO_API_KEY, also write zernio.api_key_present = true.
I already have an account: just ask ifZERNIO_API_KEYis exported.
Write zernio.signed_up = true and zernio.api_key_present accordingly.
Skip: leave defaults (false).
Phase E — First-action recommendation
Print a one-screen summary:
✓ Profile saved to ~/.claude-ads/profile.json
✓ Connected:
✓ Zernio:
Then AskUserQuestion:
> "You're set up. What would you like to run first?"
Options (filter to what the user actually configured):
/ads audit — full multi-platform audit(recommended if ≥2 platforms
connected with tier ≠ skipped).
/ads — deep-dive on one platform(single-select the
platforms they connected).
/ads plan — strategic plan from your industry template
(use their context.industry, falling back to generic).
Just exit — I'll explore on my own.
Print the EXACT next command to type (do not invoke it). Example:
> Great — next, type: /ads audit
Finalize: python3 scripts/profile.py set last_command "/ads start" && python3 scripts/profile.py set last_command_at "$(date -u +%FT%TZ)".
Status mode (/ads start status)
Run python3 scripts/profile.py get and print:
claude-ads profile
Industry:
Monthly spend: ~$/mo
Goal:
Active platforms:
Meta: · · last verified
Google: · · last verified
TikTok: · · last verified
Zernio: · API key
Last command: ()
Exit without further prompts.
Error handling
- If
python3 scripts/profile.pyis not executable, run it as `python3
scripts/profile.py …` (already the default).
- If
~/.claude-ads/is read-only, print the path and ask the user to fix
permissions before retrying.
- If
AskUserQuestionreturns "Other" without useful text, treat as cancel
for that step.
Output contract
The wizard does NOT produce a report file. It only writes ~/.claude-ads/profile.json and emits user-facing prose. Other /ads commands read the profile via python3 scripts/profile.py get.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: Hainrixz
- Source: Hainrixz/claude-ads
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.