菜单
Use when an anomaly or unusual segment/time movement appears in Atlys data and you need to explain the why by correlating it with the documented known-issues log (K1-K7) — e.g. iOS pay drop vs OTP autofill regression. Correlate, don't over-claim causation.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/sidagarwal04/click-a-thon-26-submissions --skill atlys-known-issue-correlation命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
MUST USE when reviewing ClickHouse schemas, queries, or configurations. Contains 31 rules that MUST be checked before providing recommendations. Always read relevant rule files and cite specific rules in responses.
Use this skill before running any EDA, funnel, or insight query against the Atlys `atlys` ClickHouse database, or before trusting a business/data claim about it. Loads the current business context from `agent_meta.current_context` and explains the two-database split, the context taxonomy, the size-capped ClickHouse tool pattern this repo's own agents use, and known gotchas baked into the context (funnel timestamp ordering, FX normalization, no session entity, etc.) so queries and insights are grounded instead of re-deriving (or contradicting) already-verified facts. Triggers on: "query atlys", "check the context", "what does the context say about X", "funnel analysis", "EDA on atlys", "is this consistent with base context".
Use this skill whenever you've verified something new about Atlys (a contradiction, a confirmation, a new table/entity/metric, a resolved open question) and need to write it into the living context layer — the same append-only path this repo's Chronicler agent and orchestrator use. Covers the `agent_meta.context_versions` schema, the required JSON content shape, confidence calibration, how the real orchestrator writes it (`orchestrator/pipeline.py::_write_context_sections`), and how this differs from writing to `agent_meta.insights`. Triggers on: "update the context", "fix the context", "add this finding to context", "chronicler", "record this in agent_meta", "context agent".
基于 SOC 职业分类
正在显示 SKILL.md
| name | atlys-known-issue-correlation |
| description | Use when an anomaly or unusual segment/time movement appears in Atlys data and you need to explain the why by correlating it with the documented known-issues log (K1-K7) — e.g. iOS pay drop vs OTP autofill regression. Correlate, don't over-claim causation. |
Goal: connect an observed anomaly to the documented known-issues log so insights carry the why. Correlate, never over-claim causation. Always load the current known-issues list from context (it may have grown); the entries below are the base set.
pay_now_clicked -> purchase_completed drop for os/device_type = iOS,
concentrated in Gulf geos (AE, SA, QA, KW, BH, OM).is_crossed_failed_attempt_threshold / retry_count for Android
after ~2026-04-01.retry_count correlated with citizenship of non-Latin-script
countries.destinations in
Apr–Jun. Label as seasonality.coupon_applied and lower realised
value in Q2. Signal: coupon_applied = 1 / coupon_name = 'SUMMER20' share up,
revenue per conversion down.app_version like '7.45%'.Signal columns above are named for the flat legacy tables. When the anomaly is on a newly-instrumented JSON
payloadtable, read the same signals viapayload.*— e.g.payload.os,payload.device_type,payload.`retry_count`,payload.app_version— with a CAST where needed — seeatlys-json-payload-access. The K-issue logic is identical; only the accessor changes.
"iOS conversion at the pay step is 14% below Android in AE/SA over Jun (iOS n=3.1k). This is consistent with K1 (iOS WebKit OTP autofill regression), which predicts pay abandonment for iOS Gulf card users. Confidence: Medium — direction and geo match K1, but we cannot see OTP-field events directly. Next step: instrument OTP-field focus/blur to confirm."