meta-ads-control（元广告控制）

Meta Ads Control

Use this skill for Meta Marketing API work. Prefer the bundled script because it gives structured JSON output, dry-run protection, retries, pagination, batch support, async Insights jobs, targeting search, and asset upload.

Before you touch spend or delivery

1. Discover scope.

   python3 scripts/meta_ads.py accounts
   python3 scripts/meta_ads.py account

1. Confirm the intended ad account, currency, timezone, and writable objects.
2. For any change that can spend money, change delivery, or alter tracking:

- prepare a short plan, - run --dry-run, - show the exact objects and fields to be changed, - wait for explicit user approval, - then rerun with --confirm.

1. Default new campaigns, ad sets, and ads to PAUSED unless the user explicitly asks to go live immediately.
2. After any write, do a read-after-write verification with get, list, or request GET.

Authentication and environment

The script reads:

• META_ACCESS_TOKEN — required for live API calls
• META_AD_ACCOUNT_ID — optional default account, with or without act_
• META_API_VERSION — defaults to v25.0
• META_GRAPH_BASE — defaults to https://graph.facebook.com

If META_ACCESS_TOKEN is missing, help the user set it up first instead of guessing or fabricating API responses.

OpenClaw users can inject these values through skill config. See OpenClaw notes.

Fast path by task

1) Audit or diagnose an account

Start with the smallest read that answers the question.

python3 scripts/meta_ads.py account --fields id,name,account_status,currency,timezone_name,amount_spent,spend_cap
python3 scripts/meta_ads.py list campaigns --fields id,name,objective,status,effective_status,daily_budget,lifetime_budget
python3 scripts/meta_ads.py list adsets --fields id,name,campaign_id,status,effective_status,daily_budget,lifetime_budget,optimization_goal,bid_strategy
python3 scripts/meta_ads.py list ads --fields id,name,adset_id,campaign_id,status,effective_status,creative

Use batch when you need several small reads at once. Use minimal field sets first.

2) Performance reporting

Use insights. Start with a narrow level and date window. For large windows, many fields, or breakdowns, use --async.

python3 scripts/meta_ads.py insights act_123 --level campaign --date-preset last_7d
python3 scripts/meta_ads.py insights act_123 --level ad --fields ad_id,ad_name,spend,impressions,clicks,ctr,cpc,actions,action_values,purchase_roas --date-preset last_30d --async --fetch-all

If the user asks for conversions or ROAS, include actions and action_values. If the user asks for demographic or placement splits, use --breakdowns. If they ask for action-level splits, include actions and use --action-breakdowns.

3) Pause, resume, or archive objects

Prefer set-status for single-object changes.

python3 scripts/meta_ads.py set-status 120000000000000 PAUSED --dry-run
python3 scripts/meta_ads.py set-status 120000000000000 PAUSED --confirm

For bulk operations, create a JSON batch file and use batch after approval.

4) Create or update campaign structure

Create in order:

1. campaign
2. ad set
3. creative
4. ad

Use JSON payload files for any nested params. Start from templates in assets/.

python3 scripts/meta_ads.py create campaign --params-file assets/campaign-create.json --dry-run
python3 scripts/meta_ads.py create campaign --params-file work/campaign.json --confirm

python3 scripts/meta_ads.py create adset --params-file work/adset.json --dry-run
python3 scripts/meta_ads.py create adset --params-file work/adset.json --confirm

python3 scripts/meta_ads.py create adcreative --params-file work/adcreative.json --dry-run
python3 scripts/meta_ads.py create adcreative --params-file work/adcreative.json --confirm

python3 scripts/meta_ads.py create ad --params-file work/ad.json --dry-run
python3 scripts/meta_ads.py create ad --params-file work/ad.json --confirm

If assets are local files, upload them first with upload.

5) Targeting discovery

Never invent targeting IDs. Resolve them with targeting search first.

python3 scripts/meta_ads.py targeting-search --type adinterest --q "running"
python3 scripts/meta_ads.py targeting-search --type adgeolocation --q "Munich"

Then place the returned IDs and descriptors into the ad set targeting spec.

6) Unsupported or niche endpoints

Use the low-level request subcommand.

python3 scripts/meta_ads.py request GET /act_123/reachestimate --set targeting_spec=@work/targeting.json
python3 scripts/meta_ads.py request GET /120000000000000/previews --set ad_format=DESKTOP_FEED_STANDARD

For nested values, prefer --params-file or @file.json values over many inline --sets.

Script reference

Main entry point

python3 scripts/meta_ads.py --help

Most useful subcommands

• accounts — list accessible ad accounts from the token
• account — read the default or provided account
• list — list campaigns, adsets, ads, creatives, audiences, assets, pixels, and more
• get — read a node or node edge
• create — create campaign, adset, adcreative, ad, customaudience, or any supported account edge
• update — update a node by ID
• set-status — convenience wrapper for status
• insights — sync or async reporting
• targeting-search — resolve targeting descriptors
• upload — upload to adimages or advideos
• batch — send Graph batch requests
• request — low-level escape hatch for any Graph path

Rules for good agent behaviour

• Read before write.
• Use the smallest field set that answers the question.
• Prefer JSON payload files for nested data.
• Use --fetch-all only when the user actually needs all pages.
• Use --async for heavy Insights jobs.
• Pause on repeated 613 or 80004 rate-limit errors; reduce scope or add time between retries.
• Prefer PAUSED or ARCHIVED over destructive delete operations.
• Report money in both raw API units and human units when relevant. Budgets are usually in the smallest currency denomination.
• When writing budgets, verify account currency and timezone first.
• After any mutation, verify the live state with a follow-up read.
• If the request touches housing, employment, credit, social issues, elections, or politics, check API guide for special-category cautions before proceeding.

Examples

Quick 7-day account snapshot

python3 scripts/meta_ads.py batch --batch-file assets/batch-read-example.json
python3 scripts/meta_ads.py insights act_123 --level campaign --date-preset last_7d --fields campaign_id,campaign_name,spend,impressions,clicks,ctr,cpc,actions,action_values,purchase_roas

Increase a budget safely

1. Inspect the current ad set.
2. Prepare an update payload with the new budget.
3. Dry-run it.
4. Ask for confirmation.
5. Apply and verify.

python3 scripts/meta_ads.py get 120000000000000 --fields id,name,status,effective_status,daily_budget,lifetime_budget
python3 scripts/meta_ads.py update 120000000000000 --params-file work/adset-budget.json --dry-run
python3 scripts/meta_ads.py update 120000000000000 --params-file work/adset-budget.json --confirm
python3 scripts/meta_ads.py get 120000000000000 --fields id,name,daily_budget,lifetime_budget,updated_time

Upload an image and build a link ad

python3 scripts/meta_ads.py upload adimages --file creative.jpg --confirm
# Put the returned image_hash into work/adcreative.json
python3 scripts/meta_ads.py create adcreative --params-file work/adcreative.json --confirm
python3 scripts/meta_ads.py create ad --params-file work/ad.json --confirm

References

• API guide
• Field sets and reporting defaults
• Workflow playbook
• Troubleshooting
• OpenClaw notes
• Asset templates in assets/
• Example evals in evals/evals.json