Hard verification gates that catch the claims that get published content deleted — before you hit publish. Five gates reject historical over-generalization, causal claims about tides/weather without primary data, sources that don't cover the claimed scope, stale time-sensitive facts, and one-sided reassurance. Born from real post-publication deletions on a daily knowledge channel. Works on scripts, articles, newsletters, and social posts.
Verify a versioned software release is genuinely complete — tickets actually implemented (not just closed), docs discoverable (not just present), CI green on all matrix platforms, coverage gated, security scan clean, rollback path tested, and version metadata consistent. Produces a structured READY / NOT READY / BLOCKED report with evidence-backed findings and action items.
Built around the part everyone gets wrong: when a key leaks you rotate it first — it's compromised the second it hits a remote, and bots scan public pushes within seconds. Scans staged changes, the working tree, and git history for provider key shapes, private keys, committed .env/credential files, and hardcoded creds — tuned for signal over noise so placeholders and $ENV refs don't cry wolf — then walks rotate → purge-from-history → move-to-env → add-a-pre-commit-hook, in the right order.
Expo / React Native app 上架 App Store 同 Google Play 嘅避坑清單：build 失敗、審核被拒、封測卡關、升級後相片消失 —— 四個 app 親手踩過嘅坑，逐個現象 → 原因 → 修法。
Turns a press release, official announcement, or source article into a publish-ready article while guaranteeing every date, amount, name, and location stays locked to the source — zero fabrication, verified before output.
Catch unauthorized reads, tool misuse, stale approvals, and trace-response mismatches before your AI Agent ships. Upload a trace, get evidence-backed findings, fix issues, retest, and see a clear release gate: READY, BLOCKED, or NOT_VERIFIED. Insufficient evidence may result in NOT_VERIFIED. Complete security and general task or semantic correctness are not guaranteed. Imported trace authenticity is not guaranteed.
From a local ISSUE file to a verified, committed deliverable — with a fresh-subagent acceptance gate that catches wrong screenshots and leaked logic.
收件匣一萬封的時候，「花一個週末整理」是自我安慰。 這支用 API 批次處理，三條硬原則：只搬不刪（分類錯還救得回）、每步回查伺服器端的實際計數（不信本地成功訊息）、垃圾夾預設假定是誤判——實測抽樣 77 封垃圾信裡真垃圾只有 1 封。 附八個踩坑，最值錢的是第一個：授權過期時 401 會被當成「收件匣 0 封」靜默跳過，流程回報成功但什麼都沒做。