Prompt library
Need to rename a prompt field across source, admin view models, and tests without touching generated files.
Target pattern: explicit field access in prompt source and view models. Exclusions: generated files and unrelated locale content. Review sampling: one template, one admin page, one public detail page. Formatting: run scoped formatter after codemod. Rollback: commit codemod separately from manual copy edits.
Reusable examples now come from reviewed template records, while the older catalog rows remain read-only during migration.
Producer: reviewed template records. Consumer: example aggregation and public cards. Compatibility window: older catalog rows remain read-only inventory. Validation: coverage checks and page sampling. Cleanup step: remove the old path only after final storage and page sampling pass.
Prompt detail pages return 404 for newly added slugs because static params did not include them in the build.
Impact: new prompt pages are unavailable after deploy. Suspected scope: static route generation or missing content records. Safe patch: confirm templates are included in the release and rebuild. Verification: request one new English and Chinese prompt URL. Communication: content was added but pages need a rebuild; no user data is affected.
Old prompt fixtures include database IDs, but current versioned prompts use slug as ID.
What it proves: prompt shape and required locale fields. Stale field: database ID no longer proves runtime behavior. Shared helper: build fixture from template slug and locale content. Safe deletion order: replace one fixture family, run prompts tests, then remove old IDs. Risk: admin compatibility tests may still need legacy ID examples.
Debt list: old example compatibility path, duplicate prompt scripts, long SEO titles, and missing browser sampling.
Highest priority: old example compatibility path because it affects launch example credibility. Failure risk: duplicate prompt scripts can reintroduce stale write paths. Migration pressure: final storage migration blocks release confidence. Verification cost: browser sampling is manual but bounded. Recommendation: finish storage cleanup and compatibility removal before cosmetic cleanup.
Competitor notes: three tiers, annual discount above the fold, AI credits hidden in FAQ, and customer logos near checkout. Help me find what to learn.
Positioning: the page sells lower perceived risk before it sells features. Packaging: tier names are simple, but credit limits are under-explained. Objection handling: annual savings are visible; usage anxiety is deferred to the FAQ. Trust signals: logos near checkout support the final decision moment. Tests: move credit math into the tier cards and add one buyer-specific proof point per tier.
Docs outline: setup, model selection, billing, exports, team roles. We keep getting support tickets about credits and private files.
Missing intents: credit estimation before running a job and privacy boundaries for uploaded files. Prerequisites: setup should state required account role and billing state. Stale risk: export docs need screenshots for both image and video jobs. New articles: credit planning, private file lifecycle, and team role troubleshooting. Priority: write credit planning first because it reduces pre-purchase anxiety.
Users sign up, open image generation, then leave before choosing a model. We show 18 models and no default.
Likely cause: the first decision is too wide and appears risky. Evidence to collect: model dropdown opens, hover time, failed first-run events, and search terms. Copy fix: label one default as best for product visuals and one as best for edits. Product fix: preselect a safe default and hide advanced models behind comparison. One-week experiment: default to the highest-success image model and measure first job completion.
RFP asks whether our AI workspace supports role-based access, audit logs, and customer-managed keys. We have roles and logs, CMK is planned.
Confirmed: role-based access and audit logs are available for workspace administration. Planned: customer-managed keys are on the roadmap but should not be represented as currently available. Exception: encryption details should be answered by the security owner before submission. Suggested answer: state current controls, describe the CMK roadmap carefully, and offer a security follow-up. Follow-up: ask whether CMK is mandatory for pilot approval or only for production rollout.
Clause says vendor may change AI subprocessors with notice on the website only. What should we ask?
Risk: notice may be easy to miss and may not give the team enough time to object. Business impact: privacy, procurement, and customer commitments could be affected. Question 1: can changes trigger email notice to the account owner? Question 2: is there an objection window for material subprocessor changes? Question 3: can regulated customer data be excluded from new subprocessors by default?
Objections: too many models, unclear credits, privacy questions, exports are hard to find, and team members need approvals.
Theme 1: decision overload around model choice. Theme 2: cost anxiety around credits and usage predictability. Theme 3: trust and governance concerns around privacy and approvals. Recommended reply: lead with defaults, credit estimator, and workspace controls. Product follow-up: improve model recommendation, expose export actions, and document approval flows.
Notes: onboarding templates hit the milestone, asset storage cleanup is still open, page title review has follow-ups, and usage questions continue.
Progress: template coverage reached the current target and proof depth improved. Risk: media storage cleanup remains open before launch. Decision needed: whether page title reviews should be fixed before launch or tracked as P2. Customer signal: usage confusion is still creating support volume. Next focus: storage validation, usage estimator copy, and targeted title cleanup.
Page target: AI video generator. English emphasizes cinematic clips; Chinese emphasizes prompt templates and fast export.
Intent alignment: both locales should lead with creating usable AI video, not only browsing models. English copy: keep cinematic clips, but add prompt templates and export workflow. Chinese copy: keep template speed, but add quality and controllable camera motion. Metadata: title should mention AI video generator and prompt workflow without keyword stuffing. Proof examples: use one product clip, one travel clip, and one avatar or talking-head workflow.
Change: content templates now come from files, public pages derive examples from templates, and old inline samples are compatibility data only.
Touched surfaces: content file loader, example aggregation, detail pages, and modality pages. Hidden coupling: old inline samples may still affect compatibility visuals and sitemap images. Tests: prompt template check, model content audit, route render samples, and media audit. Rollout note: treat final asset storage as a separate release gate. Watch item: any page that assumes old inline samples are the primary proof source.
We added 58 prompt templates and changed locale JSON. What regression tests should run first?
P0: prompt template schema and model category validation. P0: route rendering for one prompt page per mode. P1: SEO title and description length audit. P1: media URL existence for image, video, and audio prompts. P2: visual density checks for prompt listing filters after the count increase.
Field mediaUrl is split into imageUrl, videoUrl, audioUrl, and posterUrl. Existing clients may still send mediaUrl.
What changed: mediaUrl is now explicit by media type. Why it matters: clients can render the correct player or image component without guessing. Migration: map image assets to imageUrl, video files to videoUrl, audio files to audioUrl, and thumbnails to posterUrl. Compatibility: keep accepting mediaUrl during migration but log usage. Risk: ambiguous old values may produce incorrect previews if not mapped.
Release notes mention new default ESM loader behavior, stricter config parsing, and a changed browser revision.
Behavior changes: module loading and config validation may fail earlier. Migration work: pin loader options, update invalid config, and refresh browser caches. Tests: run typecheck, build, and at least one browser screenshot flow. Rollback signals: unexplained startup failures, config parse errors, or missing browser executable errors. Owner: platform tooling should own the upgrade and cache note.
Logs: 09:12 deploy, 09:18 media route 500s, 09:24 rollback, 09:31 traffic normal. Only prompt detail pages affected.
Timeline: deploy at 09:12, failures began at 09:18, rollback at 09:24, recovery at 09:31. Suspected trigger: media route change in the deploy. Customer impact: prompt detail pages could not load media previews for roughly 13 minutes. Mitigation: rollback restored traffic; keep the deploy frozen until route tests pass. Open questions: why prelaunch checks missed the route and whether cached pages masked the issue.
