Redesign Decodex homepage signal experience

README.md

@@ -142,7 +142,7 @@ The public site owns:
 
 - Codex signal cards
 - release-delta presentation
-- recommended config artifacts
+- continuous Radar status presentation
 - static assets and public page rendering
 
 The public site does not own:

dev/skills/README.md

@@ -21,6 +21,8 @@ follow-up work:
    `analysis_draft` JSON consumed by `scripts/github/render_signal_entry.py`.
 5. `x-post-draft`: turn evidence-backed Radar output into a reviewable
    `social_post_draft/v1` artifact for `@decodexspace`.
+6. `rate-limit-reset-watch`: review today's `@thsottiaux` X posts with AI semantic
+   judgment and refresh the homepage `reset_status/v1` artifact.
 
 Use only the skills needed for the current artifact. Do not create a social draft just
 because a signal exists.
@@ -37,6 +39,6 @@ explain what changed by themselves.
 
 Only the existing checked-in contracts are durable artifacts today:
 `github_change_bundle/v1`, `analysis_draft`, `signal_entry/v1`, `upstream_impact/v1`,
-`release_delta/v1`, and `social_post_draft/v1`. The triage, code-analysis, and
-release-analysis skills are reasoning passes unless their conclusions are promoted
-into one of those contracts.
+`release_delta/v1`, `social_post_draft/v1`, and `reset_status/v1`. The triage,
+code-analysis, release-analysis, and reset-watch skills are reasoning passes unless
+their conclusions are promoted into one of those contracts.

dev/skills/rate-limit-reset-watch/SKILL.md

@@ -0,0 +1,59 @@
+---
+name: rate-limit-reset-watch
+description: Use when checking whether today's @thsottiaux X posts semantically indicate a rate-limit reset, quota reset, message-cap recovery, or reset window, and when writing the Decodex homepage reset_status/v1 artifact under site/src/content/reset-status/.
+---
+
+# Rate Limit Reset Watch
+
+Use this repo-local skill to refresh the homepage `Are we reset today?` signal.
+The job is semantic judgment, not keyword matching.
+
+## Inputs
+
+- Source account: `https://x.com/thsottiaux`
+- Artifact: `site/src/content/reset-status/latest.json`
+- Spec: `docs/spec/reset-status.md`
+- Default timezone for "today": `Asia/Shanghai`, unless the user gives another one.
+
+## Workflow
+
+1. Determine the `observed_for_date` in the selected timezone.
+2. Collect today's visible `@thsottiaux` candidates from X.
+   - Prefer `@Chrome` when logged-in X access is needed.
+   - Use profile results and X search such as `from:thsottiaux since:YYYY-MM-DD until:YYYY-MM-DD`.
+   - If exact keyword search is useful, use it only as supporting evidence, not as the decision.
+3. Read the candidate posts, quotes, and immediate thread context needed to understand them.
+4. Decide semantically whether any candidate says rate limits reset, quota windows reset,
+   message caps recovered, or users should wait for a reset window.
+5. Write `site/src/content/reset-status/latest.json` with `schema = "reset_status/v1"`.
+6. Run the site content/type validation after updating the artifact.
+
+## Decision Rules
+
+Return `yes` when the reviewed content semantically points to reset behavior, even if the
+post never says the exact phrase `rate limit reset`.
+
+Return `no` when today's visible posts were reviewable and none are about reset behavior.
+
+Return `unknown` when X access, search, timeline loading, deleted posts, login state, or
+insufficient visible candidates make the judgment unreliable.
+
+Do not mark `yes` for generic use of words like `reset`, `limit`, `rate`, `quota`,
+or `window` unless the surrounding meaning is rate-limit reset behavior.
+
+Do not mark `yes` for unrelated posts about releases, performance, browser improvements,
+model quality, screenshots, or product announcements.
+
+## Artifact Guidance
+
+Keep `rationale` short and evidence-based. Summarize reviewed posts instead of copying
+long X text into the artifact.
+
+Use `evidence_posts[].relevance` as:
+
+- `related`: supports a `yes` judgment
+- `not_related`: reviewed but does not support reset
+- `uncertain`: could not be interpreted confidently
+
+If the artifact is `unknown`, explain the blocker in `rationale` and include any partial
+candidate summaries that were visible.

dev/skills/rate-limit-reset-watch/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Rate Limit Reset Watch"
+  short_description: "Judge whether today contains a semantic rate-limit-reset post."
+  default_prompt: "Check today’s @thsottiaux X posts and write a reset-status artifact."

docs/spec/reset-status.md

@@ -1,75 +1,86 @@
 # Reset Status
 
-Purpose: Define the runtime behavior for the homepage reset-status widget.
+Purpose: Define the artifact and homepage behavior for the reset-status widget.
 
 Status: normative
 
 Read this when:
 - You are changing the homepage reset-status widget.
-- You are changing how the client fetches OpenAI status data.
-- You need to know the current Are-we-reset heuristic.
+- You are changing how Decodex decides whether "Are we reset today?" is `Yes`, `No`, or `Unknown`.
+- You are producing or reviewing the reset-status artifact.
+
+Not this document:
+- The general social publishing workflow.
+- The Codex upstream signal-entry contract.
 
 Defines:
-- The upstream source used by the homepage widget.
-- The one-fetch-per-page-load rule.
-- The current 24-hour alert heuristic for `Yes` versus `No`.
+- The X account used as the reset-status source.
+- The `reset_status/v1` artifact consumed by the static site.
+- The AI semantic judgment rule for reset status.
 
 ## Source
 
-The widget uses the OpenAI public status page:
+The reset-status source is the X account:
+
+- account: `@thsottiaux`
+- profile URL: `https://x.com/thsottiaux`
 
-- source URL: `https://status.openai.com/incidents/01KK9JA8JKQKDW1W24T09NHBYH`
-- API URL: `https://status.openai.com/api/v2/incidents.json`
-- incident id: `01KK9JA8JKQKDW1W24T09NHBYH`
+The static site must not fetch X, OpenAI Status, or another live API from the browser for
+this widget. The site reads the latest checked-in reset-status artifact under:
 
-The homepage must fetch this API directly from the browser. Do not route this
-through a repo-owned static artifact, scheduled workflow, or generated content
-collection.
+- `site/src/content/reset-status/*.json`
 
-## Runtime fetch rule
+## Artifact
 
-The reset-status widget must fetch status data:
+The artifact schema is:
 
-- once per page open
-- from the client
-- without background polling
+- `schema`: must be `reset_status/v1`
+- `question`: must be the rendered question, currently `Are we reset today?`
+- `answer`: one of `yes`, `no`, or `unknown`
+- `confidence`: one of `confirmed`, `likely`, or `weak`
+- `observed_for_date`: the date being judged, formatted as `YYYY-MM-DD`
+- `timezone`: the timezone used to interpret "today"
+- `generated_at`: generation timestamp
+- `source_account`: the X handle that was reviewed
+- `source_url`: the reviewed profile URL
+- `search_url`: optional X search URL used during review
+- `judgment_mode`: must be `ai_semantic_review`
+- `rationale`: short explanation for the decision
+- `evidence_posts`: reviewed candidate posts or summaries, each marked as `related`,
+  `not_related`, or `uncertain`
 
-If multiple reset-status widgets are rendered on one page, they should share
-the same in-page request instead of issuing duplicate fetches.
+## Judgment Rule
 
-## Heuristic
+The reset answer is an AI semantic judgment over today's visible `@thsottiaux` posts.
+It must not be a keyword-only check.
 
-The homepage question remains:
+Set `answer` to `yes` when at least one reviewed post, quote, or directly linked context
+semantically says that rate limits reset, message caps recovered, quota windows reset, or
+users should wait for a reset window. The exact phrase `rate limit reset` is not required.
 
-- `Are we reset today?`
+Set `answer` to `no` when today's reviewed posts are sufficiently visible and none of
+them semantically indicates a rate-limit reset event.
 
-The current decision rule is:
+Set `answer` to `unknown` when X access, search, login state, timeline loading, or post
+visibility prevents a useful judgment.
 
-- `Yes` when the `Codex unresponsive` incident shows at least one alert within
-  the last `24` hours
-- `No` otherwise
+## Review Rule
 
-For this purpose, the client must first select the incident with id
-`01KK9JA8JKQKDW1W24T09NHBYH`. An alert is any incident-level or update-level
-timestamp on that incident in the last 24 hours. The client should inspect:
+The reviewer must collect today's candidate posts first, then make the semantic decision.
+Good evidence can come from X profile results, X search results, quoted context, or visible
+thread context.
 
-- `incident.created_at`
-- `incident.updated_at`
-- `incident.monitoring_at`
-- `incident.resolved_at`
-- `incident_updates[].display_at`
-- `incident_updates[].created_at`
-- `incident_updates[].updated_at`
+Do not mark `yes` from generic words such as `reset`, `limit`, `fast`, `quota`, or
+`rate` unless the surrounding context is about rate-limit reset behavior.
 
-## Failure behavior
+Do not mark `yes` from unrelated release, browser, performance, or product-quality posts.
 
-If the status fetch fails or the payload is malformed:
+## Homepage Rule
 
-- keep the widget link pointing at the OpenAI incident page
-- render the answer as `No`
-- render the widget in the muted state
+The homepage renders the checked-in artifact directly:
 
-## Source framing rule
+- `yes` renders `Yes` in the positive tone
+- `no` renders `No` in the neutral tone
+- `unknown` renders `Unknown` in the muted tone
 
-The widget remains a heuristic. It must not claim that the OpenAI status page
-directly confirms a Codex quota or usage reset event.
+The widget links to the artifact's `search_url` when present, otherwise to `source_url`.

site/.astro/content.d.ts

@@ -120,19 +120,19 @@ declare module 'astro:content' {
 		: any;
 
 	type DataEntryMap = {
-		"recommendedConfigs": Record<string, {
+		"releaseDeltas": Record<string, {
   id: string;
   body?: string;
-  collection: "recommendedConfigs";
-  data: InferEntrySchema<"recommendedConfigs">;
+  collection: "releaseDeltas";
+  data: InferEntrySchema<"releaseDeltas">;
   rendered?: RenderedContent;
   filePath?: string;
 }>;
-"releaseDeltas": Record<string, {
+"resetStatus": Record<string, {
   id: string;
   body?: string;
-  collection: "releaseDeltas";
-  data: InferEntrySchema<"releaseDeltas">;
+  collection: "resetStatus";
+  data: InferEntrySchema<"resetStatus">;
   rendered?: RenderedContent;
   filePath?: string;
 }>;