Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Crisis Playbook · minhnv0807 | DecimalAI

minhnv0807/66-crisis-playbook

SkillScore95Maintained

Playbook xu ly khung hoang truyen thong va campaign fail: phan loai 5 cap L1-L5, quy trinh 4 gio dau, template phan hoi tung loai, escalation matrix, danh sach KHONG lam, post-mortem. Dung khi nhac 'khung hoang', 'crisis', 'campaign fail', 'content gay phan ung', 'review xau lan rong', 'bi tan cong tren mang'.

marketingby minhnv0807

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Marketing & Content pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/minhnv0807-66-crisis-playbook/SKILL.md

decimalai skills pull minhnv0807-66-crisis-playbook --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/minhnv0807-66-crisis-playbook/SKILL.md https://app.decimal.ai/s/minhnv0807-66-crisis-playbook/SKILL.md

Raw file: /s/minhnv0807-66-crisis-playbook/SKILL.md ↗ · machine-readable summary at /s/minhnv0807-66-crisis-playbook.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 24d ago

Yes — +45% lift · at higher cost+91% tokens0% turns95% of whole cases pass with skill (21/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-01	✗→✓	▲ Improved	23%	0%
case-02	✗→✓	▲ Improved	16%	0%
case-03	✗→✓	▲ Improved	19%	0%
case-12	✗→✓	▲ Improved	114%	0%
case-07	✗→✓	▲ Improved	119%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/10/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

minhnv0807 / ai-business-skills

skills/vi/66-crisis-playbook/SKILL.md

View on GitHub ↗

License MITImported 8/17/2026Version v1

Skill contents

SKILL.md

Crisis Playbook

> Khung hoang den bat ngo — playbook phai co san truoc khi can dung. Ba quy tac goc: (1) khong phan ung khi chua du thong tin, phan lon crisis trong te hon thuc te trong 30 phut dau; (2) khong im lang qua 4 gio voi crisis dang lan, im lang bi hieu la thua nhan; (3) xu ly dung la co hoi xay trust, xu ly sai la mat trust vinh vien.

Thu thap thong tin

Doc 15-social-listening report neu co (phat hien som). Truoc khi lam bat cu gi, tra loi 4 cau:

Crisis loai gi? Doi chieu bang phan loai L1-L5 ben duoi.
Toc do lan? 10 nguoi biet hay 10.000 nguoi? Dang o mot kenh hay da nhay kenh?
Nguon goc tu dau? Mot khach hang khong hai long, KOL/influencer, doi thu, hay bao chi?
Su that la gi? Fact nao dang bi tranh cai, minh co bang chung khong, va co du thong tin de phan hoi chua?

Neu chua du thong tin: cong bo mot cau ngan xac nhan dang kiem tra, va hen gio cap nhat cu the. Khong im lang, cung khong phan hoi voi tin chua kiem chung.

Nguyen tac

Assess truoc, react sau. Danh 30-60 phut dieu tra con hon phan hoi sai trong 5 phut.
Khong im lang qua 4 gio voi crisis dang lan.
Khong xoa comment tieu cuc chinh dang (tru spam/hate speech ro rang) — xoa lam crisis nang them.
Khong tranh cai cong khai. Keo nguoi complaint vao kenh rieng de xu ly.
Khong phan hoi khi dang cam xuc. Viet nhap, cho 30 phut, doc lai, roi gui.
Khong hua dieu chua chac lam duoc, khong noi giam su that.
L3 tro len bat buoc bao leader, khong tu xu ly.

Phan loai crisis

| Cap | Loai | Vi du | Do uu tien | Thoi gian phan hoi | |-----|------|-------|------------|--------------------| | L1 | Operational — campaign/he thong loi, KPI miss nang | Ads khong chay, landing page down, tracking hong | High | 2-4 gio | | L2 | Content — noi dung sai su that, vo y gay hieu lam | Caption gay hieu nham, so lieu sai, hinh anh khong phu hop | High | 2-4 gio | | L3 | Reputation — complaint khach hang lan rong | Review 1 sao viral, bai boc phot | Critical | 1-2 gio | | L4 | Brand attack — tan cong co to chuc | Fake review hang loat, doi thu boi nho | Critical | Ngay | | L5 | Legal/Compliance — vi pham quy dinh, claim sai luat | Claim y te khong duoc phep, vi pham ban quyen | Critical + luat su | Ngay |

Quy trinh theo tung cap

L1 — Operational

Trong 2 gio dau: keo data day du de biet so that dang o dau; chan doan nguyen nhan goc (goi 03-danh-gia-hieu-suat); bao CEO kem hypothesis va action; trien khai fix khan.

Mau bao cao: "Campaign X] dang o so thuc] so voi target so muc tieu]. Nguyen nhan so bo: gia thuyet]. Dang lam: 3 hanh dong]. Cap nhat tiep luc gio]."

L2 — Content

Ngay lap tuc: chup man hinh luu bang chung TRUOC khi go; go bai lien quan (nhanh hon la di xin loi tren bai cu); bao leader, khong tu xu ly.

Trong 2-4 gio: dieu tra sai o dau, anh huong ai, reach bao nhieu; viet nhap phan hoi, leader duyet truoc khi dang (goi 62-marketing-review).

Mau phan hoi: "Chung toi vua phat hien noi dung X] trong bai dang ngay Y] co loi/thong tin khong chinh xac]. Chung toi da go/sua] va xin loi vi su nham lan. Thong tin dung la: ...]. Cam on nguoi phat hien] da phan anh."

L3 — Reputation

Ngay lap tuc: KHONG xoa comment, KHONG tranh cai cong khai, luu lai toan bo bang chung kem thoi gian.

Trong 1-2 gio: xac minh complaint co chinh dang khong.

Neu chinh dang: thua nhan, xin loi, sua, den bu. Mau: "Ten], chung toi rat tiec ve trai nghiem nay. Day khong phai tieu chuan chung toi muon mang den. Ban nhan tin rieng giup chung toi de xu ly truc tiep duoc khong?"
Neu khong chinh dang: trinh bay su that binh tinh, kem bang chung, khong day cang thang len.

L4 — Brand attack

Luu bang chung co thoi gian; bao CEO ngay; can nhac im lang (neu quy mo nho, phan hoi se khuech dai) hay len tieng (neu dang lan); tuyet doi khong phan hoi theo cam xuc; tham van phap ly neu co dau hieu vu khong.

L5 — Legal/Compliance

Dung toan bo campaign lien quan NGAY; bao CEO ngay; tham van luat su truoc khi lam bat cu dieu gi; khong phat ngon cong khai truoc khi co y kien phap ly.

Escalation matrix

| Cap | Ai xu ly | Ai phai biet | Timeline | |-----|----------|--------------|----------| | L1 | Media buyer + leader | CEO trong ngay | 2-4 gio | | L2 | Content + leader | CEO neu reach > 10K | 2-4 gio | | L3 | Leader | CEO ngay | 1-2 gio | | L4 | CEO + leader | Ban lanh dao neu can | Ngay | | L5 | CEO + luat su | Ban lanh dao | Ngay |

Danh sach KHONG lam

Xoa comment tieu cuc chinh dang.
Tranh cai hoac doi dau cam xuc voi nguoi dang gian.
Phan hoi khi chua kiem chung fact.
Hua den bu/xu ly dieu chua chac thuc hien duoc.
Noi doi hoac noi giam muc do nghiem trong.
Im lang qua 4 gio khi crisis dang lan.
De junior tu quyet dinh phan hoi tu cap L2 tro len.

Cau truc ket qua

markdown
# Crisis Response — [Ten su viec] — [Date/Gio]

## 1. Danh gia ban dau
| Hang muc | Noi dung |
|----------|----------|
| Phan loai | L[1-5] — [ten loai] |
| Nguon goc | |
| Quy mo hien tai | [reach uoc tinh, so kenh] |
| Fact da xac minh | |
| Fact chua xac minh | |

## 2. Response plan theo gio
| Moc thoi gian | Hanh dong | Owner | Trang thai |
|---------------|-----------|-------|------------|
| 0-1h | | | |
| 1-4h | | | |
| 4-24h | | | |
| Ngay 2-7 | | | |

## 3. Thong diep
Noi gi (da duyet): [...]
Khong noi gi: [...]
Nguoi phat ngon: [...]
Kenh phat ngon: [...]

## 4. Xu ly noi bo
| Viec | Owner | Deadline |
|------|-------|----------|

## 5. Theo doi
| Chi so | Truoc | Sau 24h | Sau 7 ngay |
|--------|-------|---------|------------|
| Sentiment | | | |
| Volume nhac ten | | | |
| Inbox/complaint | | | |

Sau khung hoang

Sau 1 tuan: viet post-mortem (chuyen gi xay ra, xu ly the nao, ket qua), cap nhat playbook voi bai hoc moi, hop team de tranh lap lai. Neu crisis lien quan campaign, gop vao 63-campaign-retrospective.

Lien ket skill

15-social-listening: phat hien som truoc khi thanh crisis; theo doi sentiment trong va sau.
03-danh-gia-hieu-suat: chan doan nguyen nhan goc cho crisis L1.
62-marketing-review: duyet noi dung phan hoi truoc khi dang.
63-campaign-retrospective: rut bai hoc he thong sau khi xu ly xong.
07-bao-cao-marketing: bao cao anh huong len KPI thang.

Checklist chat luong

] Da phan loai dung cap truoc khi phan hoi.
] Da luu bang chung co thoi gian truoc khi go bat cu thu gi.
] Fact trong phan hoi da duoc xac minh.
] Phan hoi da qua nguoi duyet (leader tro len voi L2+).
] Co nguoi phat ngon duy nhat — khong de nhieu nguoi tra loi lech nhau.
] Khong vi pham danh sach KHONG lam.
] Co lich theo doi sentiment sau 24h va 7 ngay.
] Da len lich post-mortem trong vong 1 tuan.

Illustrative Want to see what minhnv0807/66-crisis-playbook would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

9/4/2026, 09:12 UTC · 22 cases

Lift

+45%

95% vs 50%

Agent turns

1.0 vs 1.0

Tokens

+91%

4,770.09 vs 2,503

fail → pass · 11pass → pass · 10pass → fail · 1

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-13 A government regulatory body flagged our active digital ad campaign for containing an unapproved health benefit claim. The team wants to keep the ads running while drafting a compliance response over the next few days. What immediate action must be taken?	pass→pass	11,705	13,250	+13%	1	1	0%	1,764	4,577	+159%	0	0	—
case-14 Our ad tracking pixels failed during a major flash sale, causing ad spend burn without attribution tracking. What actions should be taken within the first two hours, and when must executive leadership receive an update?	pass→pass	22,694	21,305	-6%	1	1	0%	2,682	4,663	+74%	0	0	—
case-01 A popular reviewer just published a viral video exposing a severe defect in our main product and criticizing our customer service team. The video is spreading fast across social platforms. I need you to draft a Crisis Response report outlining our strategy. Structure the output into five core parts: 1) Initial incident evaluation detailing classification, origin, estimated scale, verified facts, and unverified points; 2) A timed response roadmap detailing actions across 0-1h, 1-4h, 4-24h, and days 2-7; 3) Messaging directives specifying what to say, what NOT to say, designated spokesperson, and release channels; 4) Internal task assignments with owners and deadlines; and 5) A follow-up tracking matrix monitoring sentiment, mention volume, and complaints before, at 24h, and at 7 days.	fail→pass	37,370	27,745	-26%	1	1	0%	5,142	6,322	+23%	0	0	—
case-02 Our key promotional campaign landing page went down right at the midnight launch, causing tracking errors and thousands of missed transactions while angry users are complaining on our main channel. Please create a formal crisis handling document to manage this operational failure. Make sure it provides a standard response layout: an initial situation triage table (type, source, reach, verified/unverified facts), an hourly execution breakdown covering immediate actions up to day 7, official messaging boundaries (approved content, banned topics, official spokesperson, channel), internal remediation tasks, and a metric tracking table comparing sentiment and complaint volume before the incident, after 24h, and after 7 days.	fail→pass	41,688	34,578	-17%	1	1	0%	6,072	7,039	+16%	0	0	—
case-03 We just realized our social media team posted an unverified medical assertion in a sponsored post that violates local advertising regulations, and industry competitors are calling us out publicly. Please construct a structured crisis management response plan. The report must contain: an initial assessment section evaluating the incident type, origin, scale, and fact status; a hourly response schedule spanning from 0-1h to 2-7 days; clear communication guidelines defining explicit approved statements, forbidden remarks, spokesperson, and channel; an internal action matrix with owners and deadlines; and a sentiment/mention tracking framework comparing baseline metrics against 24-hour and 7-day outcomes.	fail→pass	35,724	29,344	-18%	1	1	0%	5,182	6,161	+19%	0	0	—
case-04 We are planning a standard product feature release next month for our B2B SaaS platform. Create a 4-week proactive social media content calendar listing weekly themes, post concepts, target audience segments, and key performance indicators.	pass→fail	26,766	26,473	-1%	1	1	0%	3,554	5,807	+63%	0	0	—
case-05 Our fintech company just finalized a Series B funding round of $15 million. Please draft a formal press release to announce this funding, detailing the lead investors, growth plans, executive quotes, and boilerplate company information.	pass→pass	19,632	21,067	+7%	1	1	0%	2,042	4,792	+135%	0	0	—
case-12 An improper phrasing mistake in a social media caption went live and has reached over 25,000 views across platforms. The content writer wants to resolve this internally with their direct line manager without notifying executive leadership. Does executive leadership need to be informed?	fail→pass	18,489	17,838	-4%	1	1	0%	2,095	4,483	+114%	0	0	—
case-06 Our paid search ad campaigns on Google Ads have experienced a gradual 10% increase in cost-per-acquisition over the past month. Outline a routine performance optimization roadmap focusing on keyword negative lists, landing page A/B testing, and bidding strategy adjustments.	pass→pass	24,651	22,666	-8%	1	1	0%	3,135	5,324	+70%	0	0	—
case-07 Our marketing designer accidentally uploaded a promo poster to Facebook with an incorrect pricing discount figure. A few users noticed and commented about the error. The team wants to quickly delete the post and re-upload the corrected graphic. Provide the step-by-step response strategy for this mistake.	fail→pass	13,653	20,556	+51%	1	1	0%	2,087	4,571	+119%	0	0	—
case-08 A coordinated swarm of suspicious social media accounts is flooding our Google and Facebook pages with fake 1-star reviews and false scam accusations organized by a known rival brand. Should we start arguing back in every review thread to defend our reputation, or how should this be handled?	fail→pass	18,032	21,104	+17%	1	1	0%	2,043	4,800	+135%	0	0	—
case-09 A customer posted a viral thread 15 minutes ago alleging that our energy drink caused severe stomach illness. We have zero internal laboratory test results or batch data yet. The PR manager wants to immediately publish a detailed technical rebuttal claiming our product is 100% safe. What should our immediate response protocol be?	fail→pass	13,943	24,233	+74%	1	1	0%	2,086	5,203	+149%	0	0	—
case-10 A customer posted a viral video showing a broken seal on a premium cosmetic package they received, and the video has accumulated 50,000 views. The community manager suggests quietly deleting the video comments and blocking the customer to protect brand image. How must this complaint be handled?	pass→pass	12,378	18,564	+50%	1	1	0%	1,876	4,979	+165%	0	0	—
case-11 An online user posted a viral complaint claiming our mobile app stole funds from their account, but internal transaction logs prove conclusively that the user processed a valid self-initiated refund last week. Should we post an angry response calling them out publicly, or how should we present our response?	pass→pass	11,451	20,634	+80%	1	1	0%	1,661	4,667	+181%	0	0	—
case-15 During an ongoing public incident, our social channels are receiving dozens of dissatisfied customer comments expressing frustration with service delivery delays. The social media manager wants to purge all negative comments to keep the feed looking positive. Is this acceptable?	pass→pass	15,916	18,283	+15%	1	1	0%	1,679	4,355	+159%	0	0	—
case-16 A junior social media officer posted an inaccurate statistical graphic on our main Twitter account. Can the junior officer draft, approve, and publish the public correction statement on their own authority to save time?	pass→pass	9,721	14,809	+52%	1	1	0%	1,455	3,864	+166%	0	0	—
case-17 We just finished containing a major customer service outage crisis. When should follow-up sentiment and mention volume checks be conducted according to standard crisis protocol?	pass→pass	16,229	3,869	-76%	1	1	0%	1,688	2,929	+74%	0	0	—
case-18 Our team successfully resolved a Level 3 customer reputation issue yesterday. Should we close the ticket and move on, or is there a required post-crisis review workflow?	fail→pass	11,158	17,068	+53%	1	1	0%	1,718	4,213	+145%	0	0	—
case-19 During an active brand reputation crisis regarding product safety, multiple department leaders want to post updates from their personal social media handles to address customer questions faster. Is multi-person public messaging recommended?	pass→pass	16,877	16,594	-2%	1	1	0%	1,728	4,010	+132%	0	0	—
case-20 A negative complaint about product quality is rapidly gaining traction across online forums. The PR team wants to delay any public acknowledgement for 12 hours while waiting for a full board meeting tomorrow morning. Is this timeline compliant?	fail→pass	19,414	12,782	-34%	1	1	0%	2,178	4,233	+94%	0	0	—
case-21 An angry influencer posted an aggressive tweet attacking our team. The PR specialist typed an urgent draft response while feeling agitated and wants to publish it immediately. What cooling-off protocol must be followed?	fail→pass	15,856	16,380	+3%	1	1	0%	1,677	4,180	+149%	0	0	—
case-22 A viral customer complaint post regarding defective goods is currently accumulating hundreds of shares per hour. What is the required initial response time window for handling this level of reputation crisis?	fail→pass	15,128	14,566	-4%	1	1	0%	1,522	3,770	+148%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +45 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 50%
Pass rate with the skill: 95.5%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 1
Skill version measured: v1
Run completed: 9/4/2026, 09:12 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-01

Prompt

A popular reviewer just published a viral video exposing a severe defect in our main product and criticizing our customer service team. The video is spreading fast across social platforms. I need you to draft a Crisis Response report outlining our strategy. Structure the output into five core parts: 1) Initial incident evaluation detailing classification, origin, estimated scale, verified facts, and unverified points; 2) A timed response roadmap detailing actions across 0-1h, 1-4h, 4-24h, and days 2-7; 3) Messaging directives specifying what to say, what NOT to say, designated spokesperson, and release channels; 4) Internal task assignments with owners and deadlines; and 5) A follow-up tracking matrix monitoring sentiment, mention volume, and complaints before, at 24h, and at 7 days.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run95

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+124% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 23 days ago

Other measured skills in the registry, with their headline benchmark lift.