Automating Global Release Notes
Building a Claude Cowork plugin that ships app store copy for 30 languages every week and learns from reviewer corrections.
Content design (sole automation builder)
Claude Cowork, Lark Sheets API, Lark Bitable
4 weeks to production, 6 weeks to v0.8.3 (latest)
3–4 hours manual work reduced to 10 minutes
Automating Global Release Notes
Building a Claude Cowork plugin that ships app store copy for 30 languages every week and learns from reviewer corrections.
Content design (sole automation builder)
Claude Cowork, Lark Sheets API, Lark Bitable
4 weeks to production, 6 weeks to v0.8.3 (latest)
3–4 hours manual work reduced to 10 minutes
1: The Catalyst
The legacy release notes pipeline relied on manual tracking with near 0% cross-functional compliance, forcing me into an inefficient, reactive auditing loop. This structural vulnerability triggered a major P0 compliance leak during the high-stakes launch of the OKX debit card.
Because the tracking sheets lacked regional targeting parameters, market-restricted product copy leaked live onto the United States storefront, a market where OKX Card was legally and operationally unavailable. This critical failure became my direct catalyst to build an automated, region-locked release engine.
2: The Technical Evolution (Chrome Agent vs. API Pivot)
The first version of the automation tried to mimic a human operator by using a script to take screenshots of the tracking sheets and read the text fields. It was incredibly slow, took over an hour to run and broke constantly.
The breakthrough was switching to a direct data connection using the Lark Sheets API, once made available. Instead of a digital assistant manually clicking around a browser screen, the engine then updated all 30 language rows simultaneously in just a few seconds. The system only loads a browser at the very end to automatically drop comments tagging specific regional language managers whenever they need to review a custom entry.
3: Core Automation Logic
For weeks without new features, the engine simply pulled from an approved bank of generic release notes to publish the update automatically. For weeks with active product updates, the system drafted the custom text and automatically dropped comment tags directly on the tracking sheet. This instantly alerted the exact regional language managers who needed to log in and review those specific custom entries before anything went live.
4: System Guardrails & Impact
Shifting to this programmatic pipeline turned a fragmented manual content pull into a 10-minute automated run, completely eliminating the cross-border communication lag that previously stalled deployments across international time zones.
Beyond raw efficiency, the engine was designed as a live quality gate for global compliance. Because AI language models inherently risk text hallucinations, the system utilized an automated semantic consistency gate to intercept any flagged content before going live.
The framework successfully caught and rewrote 3 high-risk storefront leaks where generated translations in Russian, Turkish and Portuguese included hallucinated features and/or restricted product terms, automatically halting the deployment loop and routing the cells to local language managers for human audit before any errors could be published to app stores.
1: The Catalyst
The legacy release notes pipeline relied on manual tracking with near 0% cross-functional compliance, forcing me into an inefficient, reactive auditing loop. This structural vulnerability triggered a major P0 compliance leak during the high-stakes launch of the OKX debit card.
Because the tracking sheets lacked regional targeting parameters, market-restricted product copy leaked live onto the United States storefront, a market where OKX Card was legally and operationally unavailable. This critical failure became my direct catalyst to build an automated, region-locked release engine.
2: The Technical Evolution (Chrome Agent vs. API Pivot)
The first version of the automation tried to mimic a human operator by using a script to take screenshots of the tracking sheets and read the text fields. It was incredibly slow, took over an hour to run and broke constantly.
The breakthrough was switching to a direct data connection using the Lark Sheets API, once made available. Instead of a digital assistant manually clicking around a browser screen, the engine then updated all 30 language rows simultaneously in just a few seconds. The system only loads a browser at the very end to automatically drop comments tagging specific regional language managers whenever they need to review a custom entry.
3: Core Automation Logic
For weeks without new features, the engine simply pulled from an approved bank of generic release notes to publish the update automatically. For weeks with active product updates, the system drafted the custom text and automatically dropped comment tags directly on the tracking sheet. This instantly alerted the exact regional language managers who needed to log in and review those specific custom entries before anything went live.
4: System Guardrails & Impact
Shifting to this programmatic pipeline turned a fragmented manual content pull into a 10-minute automated run, completely eliminating the cross-border communication lag that previously stalled deployments across international time zones.
Beyond raw efficiency, the engine was designed as a live quality gate for global compliance. Because AI language models inherently risk text hallucinations, the system utilized an automated semantic consistency gate to intercept any flagged content before going live.
The framework successfully caught and rewrote 3 high-risk storefront leaks where generated translations in Russian, Turkish and Portuguese included hallucinated features and/or restricted product terms, automatically halting the deployment loop and routing the cells to local language managers for human audit before any errors could be published to app stores.