amazon-ads-manager

Manage Amazon Advertising campaigns via the official Advertising API. Read live campaign/keyword/search-term performance, calculate ACOS/ROAS/CTR, identify wasted spend, adjust bids, pause keywords, add negatives. Trigger on: 广告优化, 广告分析, ACOS, 竞价, 投放, 关键词出价, 广告报告, 浪费花费, campaign analysis, ads optimization, keyword bid, sponsored products, 广告表现, 哪些词在烧钱, adjust bid, pause campaign.

Amazon Advertising API wrapper for live campaign management — read performance reports, update bids, pause keywords, add negatives.

Setup (one-time)

Create a .env file in the skill root directory (same level as this SKILL.md):

text

AMAZON_ADS_CLIENT_ID=amzn1.application-oa2-client.xxx
AMAZON_ADS_CLIENT_SECRET=xxx
AMAZON_ADS_REFRESH_TOKEN=Atzr|xxx
AMAZON_ADS_PROFILE_ID=          # find via `ads.py profiles`
AMAZON_ADS_REGION=NA            # NA | EU | FE

How to get credentials:

1. Amazon Ads Developer Console → create app → get client_id + client_secret
2. OAuth flow → get refresh_token (login with your Seller Central account)
3. Run ads.py profiles to find your AMAZON_ADS_PROFILE_ID (one per marketplace)

If credentials are not yet configured: ask the skill owner (the person who installed this skill) to provide the Amazon Ads API credentials and save them to the .env file listed above. Without these, all commands will fail with a clear error message listing exactly which variables are missing.

---

Commands

List profiles (find your PROFILE_ID)

bash

uv run <skill-dir>/scripts/ads.py profiles

Returns all linked ad accounts with profileId, marketplace, currency. Set AMAZON_ADS_PROFILE_ID to the correct profileId for subsequent calls.

---

View campaigns

bash

# Active campaigns
uv run ... campaigns

# All (including paused)
uv run ... campaigns --state all

Returns: campaignId, name, state, budget, bidding strategy.

---

View ad groups

bash

uv run ... adgroups <campaign_id>

View keywords

bash

uv run ... keywords <adgroup_id>

Returns: keywordId, keywordText, matchType, state, bid.

---

Performance reports (async, ~30–90 sec)

bash

# Campaign-level ACOS/ROAS summary (last 30 days)
uv run ... report campaigns --days 30

# Keyword-level performance
uv run ... report keywords --days 30

# Search terms (harvesting + negation)
uv run ... report searchterms --days 30

# Scope to one campaign
uv run ... report keywords --days 30 --campaign-id 12345678

Output fields (enriched): impressions, clicks, cost, attributedSales7d, attributedUnitsOrdered7d, acos_pct, roas, ctr_pct, cpc

Sorted by spend descending.

---

Update keyword bid

bash

uv run ... set-bid <keyword_id> <new_bid>
# e.g.
uv run ... set-bid 987654321 0.85

Pause / enable / archive keyword

bash

uv run ... set-state <keyword_id> paused
uv run ... set-state <keyword_id> enabled

Add campaign-level negative keyword

bash

# Negative exact (default)
uv run ... add-negative <campaign_id> "bad search term"

# Negative phrase
uv run ... add-negative <campaign_id> "irrelevant" --match phrase

---

Optimization Workflow

Step 1 — Get overview

bash

uv run ... report campaigns --days 30

Flag campaigns with acos_pct > 50% for drill-down.

Step 2 — Drill into problem campaigns

bash

uv run ... report keywords --days 30 --campaign-id <id>

Classify each keyword:

Condition	Action
spend > $5, sales = 0, clicks ≥ 10	🔴 Pause — pure waste
acos_pct > target × 1.5, clicks ≥ 20	🟠 Reduce bid by 20–30%
acos_pct > 0, acos_pct < target, impressions < 500	🟡 Raise bid by 20%
ctr_pct > 0.5%, orders/clicks < 5%	🟡 Listing problem — don't touch bid
impressions = 0	Check match type / bid floor

Target ACOS benchmarks: SP Manual 25–35% · SP Auto 35–45% · SB 40–55%

Step 3 — Harvest search terms

bash

uv run ... report searchterms --days 30

From auto campaigns:

• acos_pct < target AND clicks ≥ 5 AND sales > 0 → add to manual exact at 80% of current auto bid
• spend > $3 AND sales = 0 AND clicks ≥ 8 → add as negative exact

Step 4 — Execute changes

For each action identified, run the appropriate command (set-bid, set-state, add-negative). Present a summary of what was changed and the expected impact.

---

Output Format

After analysis, present:

text

## 广告优化报告

### 账户概览 (最近 N 天)
| 总花费 | 总销售额 | ACOS | ROAS | 点击数 | 平均 CPC |
|--------|---------|------|------|--------|---------|

### 🔴 高优先级 — 建议暂停 (X 个词，节省 $X/月)
| Campaign | Ad Group | Keyword | 花费 | 点击 | 销售 | 操作 |

### 🟠 降价建议 (X 个词)
| Keyword | 当前 bid | 建议 bid | 当前 ACOS | 目标 ACOS |

### 🟢 收割词 — 建议加入 Manual (X 个词)
| 搜索词 | Auto 花费 | ACOS | 建议 bid |

### 执行摘要
已执行 / 等待确认 X 项操作

Always ask for confirmation before executing bid changes or pauses unless the user explicitly said "直接执行" or "just do it".