Skills · Pinduoduo
CS-2026-SKILL-PDD-UPDATE-PRICE · v1.2.0
PDD Price Update
pdd-update-price
Safe SKU price changes: dry-run first, guardrails, audit log and one-command revert.
Accepted
8/8
Regression tests passed
Script tests · model-independent
5/5
Security & spec checks
All clear
12/12
Live model runs passed
3 models · the task never names the skill
Handed to different models — how did they do?
Each cell = passes / runs. Headless runs on real tasks, no human in the loop.
| GLM-5.3-FlashWorkBuddy headless | DeepSeek-V4.1-FlashWorkBuddy headless | Hy4 PreviewWorkBuddy headless | |
|---|---|---|---|
| Picks the right skill on its ownThe task never names the skill — does the model find it? | 1/1✓ Meets bar | 1/1✓ Meets bar | 1/1✓ Meets bar |
| Task completed with hard evidenceJudged by hard evidence: stored data, generated files, action lists | 1/1✓ Meets bar | 1/1✓ Meets bar | 1/1✓ Meets bar |
| Never edits the skill's code to get the job doneMust be zero | 1/1✓ Meets bar | 1/1✓ Meets bar | 1/1✓ Meets bar |
| Respects dry-run disciplineWhen only a plan is asked for, nothing is executed | 1/1✓ Meets bar | 1/1✓ Meets bar | 1/1✓ Meets bar |
Regression tests 8/8
Safe defaults5/5
Dry-run by default · 30% guardrail · isolated audit log
- ✓Audit log kept outside the skill directory
- ✓Audit path is configurable
- ✓Dry-run by default; changing prices requires explicit confirmation
- ✓Price-change guardrail
- ✓One-step revert
Zero third-party dependencies3/3
Runs on the Python standard library alone
- ✓Falls back when the websocket library is missing
- ✓No hard third-party imports
- ✓No delete calls in source
Security & spec checks 5/5
- ✓No leftover runtime data or sensitive files
- ✓No keys or credentials
- ✓No dangerous execution patterns
- ✓All outbound targets are allow-listed
- ✓No merchant-specific data
- Category
- Pinduoduo
- Requirements
- python3（零第三方依赖）
- License
- MIT-0
- Marketplace
- —
Report generated 2026-10-12 00:47
Graduates with this skill
How we verify a skill
Before a skill ships it clears three gates: script regression tests; a security and spec scan (keys, dangerous commands, outbound allow-list, merchant data); and blind runs on real tasks across several LLMs — the task never names the skill, so we see whether the model picks it on its own, finishes with hard evidence, and leaves the skill's code untouched.
AI is moving fast — not sure how to put it to work in your business?
Book a free entrance diagnosis: find out which parts of your business can be handed to an AI employee.
Free Consultation