just-vibe 0.7.0

package/plugins/just-vibe/skills/ui-responsive/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: ui-responsive
+description: "Fix layouts across screen sizes and input methods Use for layout adaptation, zoom and input differences; responsive is the same canonical workflow."
+---
+
+# ui-responsive
+
+Fix layouts across screen sizes and input methods
+
+## Choose this workflow
+
+Use for layout adaptation, zoom and input differences; responsive is the same canonical workflow.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [UI and frontend methods](../../references/packs/ui.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **apply**. Apply; target layouts, content extremes, and supported input/viewport conditions.
+
+target screens/flows, existing design conventions, and runnable UI or supplied references. Visual claims require actual renders; accessibility claims distinguish automated, keyboard, and assistive-technology evidence.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Layout adaptation including touch, pointer, zoom, and keyboard effects; shares primitives with `responsive`.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Inspect intrinsic sizes and flow, reproduce failures, adjust layout constraints, and verify intermediate widths, long text, and relevant orientation changes.
+- Find the intrinsic width constraint or overflow source, adjust layout at content-driven boundaries and test nearby widths with long text and keyboard focus.
+
+## Decision branches
+
+- **When hiding an element would remove required functionality:** Reflow or provide an equivalent reachable interaction instead of suppressing it.
+
+## Deliver and verify
+
+- Layout fixes and a viewport/input coverage record.
+- Viewport/content/input matrix and observed reachability/overflow results.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Content remains reachable at narrow widths and zoom; hover-only controls have a usable alternate interaction.
+
+## Stop and recover
+
+- Do not hide required functionality or claim device coverage from a single desktop screenshot.
+
+## Example requests
+
+- **Normal (apply):** Fix the checkout layout for narrow screens, zoom, and touch input.
+- **edge (apply):** Repair a table at narrow widths and high zoom without hiding essential actions.
+- **blocked (inspect):** Review responsive source and screenshots without claiming real-device interaction coverage.

package/plugins/just-vibe/skills/ui-states/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: ui-states
+description: "Add loading, empty, error, partial-data, disabled, and success states Use to implement missing interface states; ui-flow connects states across a journey."
+---
+
+# ui-states
+
+Add loading, empty, error, partial-data, disabled, and success states
+
+## Choose this workflow
+
+Use to implement missing interface states; ui-flow connects states across a journey.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [UI and frontend methods](../../references/packs/ui.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **apply**. Apply; feature and valid loading/empty/error/partial/disabled/success conditions.
+
+target screens/flows, existing design conventions, and runnable UI or supplied references. Visual claims require actual renders; accessibility claims distinguish automated, keyboard, and assistive-technology evidence.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Missing state behavior and presentation for the selected feature.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Derive states from data and business rules, define transitions and recovery, implement accessible feedback, and exercise each state with controlled data.
+- Derive loading, empty, partial, stale, failed and success states from the data contract; define retry and back-navigation transitions before rendering them.
+
+## Decision branches
+
+- **When a partial response has useful data and an error:** Preserve usable content and explain the failed portion rather than displaying a misleading empty state.
+
+## Deliver and verify
+
+- State-complete UI plus transition/state verification.
+- State/trigger/message/action table and controlled fixture checks.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Empty data is distinguished from failed loading; a retry preserves relevant user input and context.
+
+## Stop and recover
+
+- Do not invent payment/account policies. Unspecified critical transitions become questions while independent states proceed.
+
+## Example requests
+
+- **Normal (apply):** Complete billing states for missing cards, failed renewal, and successful recovery.
+- **edge (apply):** Add partial failure and retry states to a dashboard without losing filters.
+- **blocked (inspect):** Design states with unknown payment policy; leave the policy-dependent transition unresolved.

package/plugins/just-vibe/skills/ui-system/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: ui-system
+description: "Establish typography, spacing, colors, tokens, and component conventions Use to establish or refine shared design tokens/components; polish makes local refinements."
+---
+
+# ui-system
+
+Establish typography, spacing, colors, tokens, and component conventions
+
+## Choose this workflow
+
+Use to establish or refine shared design tokens/components; polish makes local refinements.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [UI and frontend methods](../../references/packs/ui.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **plan**. Plan; existing screens, brand constraints, reusable components, and desired consistency.
+
+target screens/flows, existing design conventions, and runnable UI or supplied references. Visual claims require actual renders; accessibility claims distinguish automated, keyboard, and assistive-technology evidence.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Tokens and component conventions; implementation requires an explicit build/adoption request.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Inventory existing values, identify a coherent scale, define semantic tokens and component states, and plan incremental adoption without visual regressions.
+- Inventory actual repeated values and component states; separate semantic roles from raw scales and define incremental adoption with representative specimens.
+
+## Decision branches
+
+- **When two themes require different contrast relationships:** Map semantic tokens per theme and verify components rather than applying one global color substitution.
+
+## Deliver and verify
+
+- Typography/spacing/color/state system and migration mapping or authorized implementation.
+- Token roles/scales, component state matrix and migration examples.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Tokens express meaning across components; contrast and long-content behavior remain usable in representative states.
+
+## Stop and recover
+
+- Do not replace branding or add a component framework without need. Resolve competing theme requirements explicitly.
+
+## Example requests
+
+- **Normal (plan):** Plan semantic design tokens from the existing screens and brand constraints.
+- **edge (plan):** Consolidate spacing and color tokens across light and dark settings screens.
+- **blocked (inspect):** Plan a system from existing UI without replacing unavailable brand assets.

package/plugins/just-vibe/skills/ui-visual-diff/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: ui-visual-diff
+description: "Compare screenshots against an accepted reference Use to compare matched visual states; ui-audit judges usability of the resulting interface."
+---
+
+# ui-visual-diff
+
+Compare screenshots against an accepted reference
+
+## Choose this workflow
+
+Use to compare matched visual states; ui-audit judges usability of the resulting interface.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [UI and frontend methods](../../references/packs/ui.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **inspect**. Inspect existing captures; reference/current renders with matching viewport, state, data, fonts, and theme.
+
+target screens/flows, existing design conventions, and runnable UI or supplied references. Visual claims require actual renders; accessibility claims distinguish automated, keyboard, and assistive-technology evidence.
+
+Declared evidence requirements: `browser.inspect`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Visual differences and regression classification; automatic fixes belong to `match` or an explicit apply request.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Normalize capture conditions, identify meaningful regions, separate dynamic content noise, and describe differences by user impact.
+- Match viewport, DPR, fonts, content, theme and animation state; mask only justified nondeterminism and inspect changed regions before accepting baselines.
+
+## Decision branches
+
+- **When captures differ in environment or content:** Recreate comparable captures or give qualitative differences without a precise pixel claim.
+
+## Deliver and verify
+
+- Compared images or annotated difference report with conditions and confidence.
+- Capture conditions, changed regions, accepted differences and remaining mismatches.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A layout shift is detected; a timestamp-only change is not misclassified as a design regression.
+
+## Stop and recover
+
+- New captures require authorized browser execution. Unmatched conditions invalidate precise pixel-difference claims.
+
+## Example requests
+
+- **Normal (inspect):** Compare these matched-viewport screenshots and separate dynamic noise from regressions.
+- **edge (inspect):** Compare screenshots with dynamic timestamps and a real layout shift.
+- **blocked (inspect):** Compare supplied captures with unknown font loading; do not assert exact pixel fidelity.

package/plugins/just-vibe/skills/vercel-audit/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: vercel-audit
+description: "Inspect project configuration, build settings, and deployment assumptions Use to compare repository configuration with a specific Vercel project; vercel-runtime investigates a particular runtime failure."
+---
+
+# vercel-audit
+
+Inspect project configuration, build settings, and deployment assumptions
+
+## Choose this workflow
+
+Use to compare repository configuration with a specific Vercel project; vercel-runtime investigates a particular runtime failure.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Vercel methods](../../references/packs/vercel.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **inspect**. Inspect; project and repository plus reported deployment concerns.
+
+exact team/project/environment and deployment/revision when applicable; read access to relevant configuration/logs. Verify installed CLI/API support and framework behavior during implementation. Never print environment values or infer promotion authorization from a preview request.
+
+Declared evidence requirements: `vercel.context`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Framework detection, root/build/output settings, runtime assumptions, and deployment configuration.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Compare repository scripts/configuration with project settings, inspect recent deployment metadata, and identify drift or unsupported assumptions.
+- Record team/project/revision and compare root directory, build/install command, output directory, framework preset and runtime against the relevant package.
+
+## Decision branches
+
+- **When a monorepo's deployed root differs from the package under review:** Trace install/build working directories and workspace dependency resolution before changing settings.
+
+## Deliver and verify
+
+- Configuration inventory, evidence-backed findings, and ordered fixes.
+- Setting/source/effective-value comparison with drift and unavailable observations.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A wrong monorepo root is identified; missing settings access is reported rather than assumed to match local defaults.
+
+## Stop and recover
+
+- No settings edits or new deployment. Do not treat a successful old deployment as evidence the current revision is healthy.
+
+## Example requests
+
+- **Normal (inspect):** Audit the linked project's build settings against this monorepo configuration.
+- **edge (inspect):** Audit a monorepo deploying the wrong workspace package.
+- **blocked (inspect):** Audit supplied settings and logs without live Vercel access.

package/plugins/just-vibe/skills/vercel-build-fix/SKILL.md

@@ -0,0 +1,63 @@
+---
+name: vercel-build-fix
+description: "Reproduce and repair failed deployment builds Use for a failed Vercel build; vite-bundle handles size and splitting of a successful build."
+---
+
+# vercel-build-fix
+
+Reproduce and repair failed deployment builds
+
+## Choose this workflow
+
+Use for a failed Vercel build; vite-bundle handles size and splitting of a successful build.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Vercel methods](../../references/packs/vercel.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **apply**. Apply; failed deployment, branch/revision, and build logs.
+
+exact team/project/environment and deployment/revision when applicable; read access to relevant configuration/logs. Verify installed CLI/API support and framework behavior during implementation. Never print environment values or infer promotion authorization from a preview request.
+
+Declared evidence requirements: `vercel.context`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Repository/build configuration causing the failure; project settings changes require explicit target scope.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Identify the failing deployment, SHA, environment and first causal build error. Compare repository/install root, build package, output path, runtime/package manager and resolved dependencies with the successful environment.
+- Reproduce the failing boundary locally when possible using the same workspace command and versions. Check case sensitivity, hoisted undeclared dependencies, build-time environment names and generated-file assumptions before patching application behavior.
+- Apply a focused fix and verify the corresponding build path. Keep local success separate from remote deployment verification; reuse the existing deployment/project identity and create a new deployment only when requested.
+- Report the causal evidence, changed configuration/code, local result and the deployment/revision actually observed remotely. Redact values and do not download secrets as incidental diagnosis.
+- Use the matching bundled evidence collector when available; read its result and limitations rather than treating exit zero as readiness. Revalidate identity before a dependent action.
+
+## Read when relevant
+
+- Resolving check, deployment or migration identity: [Delivery evidence](../../references/scenarios/delivery-evidence.md).
+
+## Decision branches
+
+- **When local build succeeds but deployment fails:** Reproduce the specific environment difference before changing application code or adding dependencies.
+
+## Deliver and verify
+
+- Fix, root-cause explanation, local check results, and deployment verification if authorized.
+- Causal build boundary, environment comparison, patch and reproduction outcome.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A missing runtime dependency is resolved correctly; environment differences are not masked by hard-coded secret values.
+
+## Stop and recover
+
+- Do not deploy or alter production settings merely to test. Inaccessible logs limit the diagnosis explicitly.
+
+## Example requests
+
+- **Normal (apply):** Repair the preview build failure using the supplied deployment logs.
+- **edge (apply):** Fix a deployment where an omitted runtime dependency exists only through workspace hoisting.
+- **blocked (inspect):** Diagnose from a build log without a valid Vercel token; do not deploy to test.

package/plugins/just-vibe/skills/vercel-env/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: vercel-env
+description: "Compare required variable names and scopes without exposing values Use to inspect or explicitly manage variable names/scopes; vite-env traces client exposure and build modes."
+---
+
+# vercel-env
+
+Compare required variable names and scopes without exposing values
+
+## Choose this workflow
+
+Use to inspect or explicitly manage variable names/scopes; vite-env traces client exposure and build modes.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Vercel methods](../../references/packs/vercel.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **inspect**. Inspect; project, environment/branch scopes, and required variable names.
+
+exact team/project/environment and deployment/revision when applicable; read access to relevant configuration/logs. Verify installed CLI/API support and framework behavior during implementation. Never print environment values or infer promotion authorization from a preview request.
+
+Declared evidence requirements: `vercel.context`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Presence, scope, and source of configuration names; values remain secret.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Derive required names from code/configuration, compare authorized metadata, identify missing/mis-scoped variables, and explain rebuild/redeployment implications.
+- Map each referenced name to build-time or runtime usage and intended environment/branch; report presence and scope without fetching secret values.
+
+## Decision branches
+
+- **When a changed value is compiled into a static client bundle:** Explain the required rebuild/deployment and inspect public exposure; editing a variable alone does not change an existing bundle.
+
+## Deliver and verify
+
+- Redacted name/scope matrix and proposed corrections.
+- Variable-name table with consumer, environment, branch scope and rebuild implications.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A preview-only missing variable is distinguished from production; reports and commands never include secret values.
+
+## Stop and recover
+
+- Mutation needs an explicit configuration request and secure input channel. Do not fetch values when metadata suffices or copy production secrets into previews.
+
+## Example requests
+
+- **Normal (inspect):** Compare required variable names across preview and production without exposing values.
+- **edge (inspect):** Inspect a variable present in production but absent on a branch-specific preview.
+- **blocked (inspect):** Assess required environment names from source without downloading secret values.

package/plugins/just-vibe/skills/vercel-performance/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: vercel-performance
+description: "Investigate slow routes using available measurements and logs Use for measured deployment latency/cache problems; react-rerenders handles client render cost."
+---
+
+# vercel-performance
+
+Investigate slow routes using available measurements and logs
+
+## Choose this workflow
+
+Use for measured deployment latency/cache problems; react-rerenders handles client render cost.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Vercel methods](../../references/packs/vercel.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **inspect**. Inspect; slow routes, deployment, workload, and existing measurements.
+
+exact team/project/environment and deployment/revision when applicable; read access to relevant configuration/logs. Verify installed CLI/API support and framework behavior during implementation. Never print environment values or infer promotion authorization from a preview request.
+
+Declared evidence requirements: `vercel.context`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Server response, cold starts, caching, payloads, and relevant client delivery behavior.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Correlate timing with runtime/cache state, separate server from network/client delays, compare like-for-like requests, and rank optimizations by evidence.
+- Separate cold start, warm handler, dependency, network and browser timing; compare matching regions, payloads and cache states.
+
+## Decision branches
+
+- **When a fast sample is cached and a slow sample is uncached:** Report separate distributions and investigate cache eligibility before claiming compute regression.
+
+## Deliver and verify
+
+- Bottleneck analysis, measurement conditions, and bounded optimization plan.
+- Measurement conditions, distributions, limiting boundary and bounded optimization experiment.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Cache hits and misses are compared separately; a single slow sample does not become a universal latency claim.
+
+## Stop and recover
+
+- New load/profiling requires authorized execution and budget. No automatic paid plan upgrade or unrelated application rewrite.
+
+## Example requests
+
+- **Normal (inspect):** Analyze these route timings; distinguish cached responses from cold execution.
+- **edge (inspect):** Compare slow preview requests with cached production responses fairly.
+- **blocked (inspect):** Assess supplied timing samples without load testing or changing the paid plan.

package/plugins/just-vibe/skills/vercel-preview/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: vercel-preview
+description: "Prepare and validate a branch or PR preview deployment Use for a specifically requested preview deployment; vercel-release-check assesses production readiness."
+---
+
+# vercel-preview
+
+Prepare and validate a branch or PR preview deployment
+
+## Choose this workflow
+
+Use for a specifically requested preview deployment; vercel-release-check assesses production readiness.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Vercel methods](../../references/packs/vercel.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **plan**. Plan; branch/PR, project, revision, and preview goal. An explicit create-preview request authorizes that preview deployment.
+
+exact team/project/environment and deployment/revision when applicable; read access to relevant configuration/logs. Verify installed CLI/API support and framework behavior during implementation. Never print environment values or infer promotion authorization from a preview request.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Preview environment only, including authorized smoke checks.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Resolve project/revision, inspect prerequisites, reuse a matching deployment where suitable, create only when requested, and verify URL, revision, routing, and key behavior.
+- Resolve intended commit and project, reuse a matching deployment if appropriate and verify identity, access protection, routes and representative behavior.
+
+## Decision branches
+
+- **When a deployment request times out:** Query for the intended revision before retrying; distinguish a protected URL from an unhealthy app.
+
+## Deliver and verify
+
+- Preview plan or actual URL/deployment ID with smoke-check results.
+- Preview identity/URL, exact revision, access requirements and observed checks.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- The preview serves the intended commit; a protected preview is reported as access-restricted rather than broken without evidence.
+
+## Stop and recover
+
+- No promotion, production-domain changes, or unsolicited sharing. Check for existing deployment after uncertain creation.
+
+## Example requests
+
+- **Normal (plan):** Prepare a preview for this branch and project; identify prerequisites first.
+- **edge (plan):** Create a preview after an earlier request returned no deployment ID.
+- **blocked (inspect):** Plan a preview without provider access; do not invent a URL.

package/plugins/just-vibe/skills/vercel-release-check/SKILL.md

@@ -0,0 +1,57 @@
+---
+name: vercel-release-check
+description: "Verify a deployment and prepare promotion or rollback steps Use to verify a named deployment's release gates; deploy executes authorized transition."
+---
+
+# vercel-release-check
+
+Verify a deployment and prepare promotion or rollback steps
+
+## Choose this workflow
+
+Use to verify a named deployment's release gates; deploy executes authorized transition.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Vercel methods](../../references/packs/vercel.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **inspect**. Inspect; candidate deployment, intended production target, health criteria, and previous stable deployment.
+
+exact team/project/environment and deployment/revision when applicable; read access to relevant configuration/logs. Verify installed CLI/API support and framework behavior during implementation. Never print environment values or infer promotion authorization from a preview request.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Readiness and promotion/rollback preparation; actual promotion is a separate authorized action.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Resolve team/project, candidate deployment ID, immutable Git SHA and intended environment from observed metadata. Compare that identity with the tested commit, build output and required configuration names/scopes.
+- Check build completion, representative route/API health, asset content types, access protection and environment-specific behavior. Distinguish a protected preview from an unhealthy deployment; do not use a generic successful homepage as proof of the application journey.
+- Identify the previous known-good deployment and test the proposed recovery against schema/data and external effects. Reverting code may not restore data compatibility or undo published messages.
+- Produce a gate table tied to this candidate: criterion, evidence identity/time, pass/fail/unknown and blocking consequence. Re-read candidate identity before readiness is reported. Promotion, alias/DNS changes and rollback require their own authorized action/target.
+
+## Decision branches
+
+- **When rollback would restore code but not reverse a schema/data change:** Mark that recovery gap and require an explicit compatible recovery plan.
+
+## Deliver and verify
+
+- Candidate identity, evidence-linked release gates, concrete blockers or unknowns, recovery target/limits and readiness for the exact requested action.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- All readiness evidence applies to the named deployment and revision; unknown health or incompatible recovery is not converted to a pass. No deployment or promotion is implied by an audit.
+
+## Stop and recover
+
+- Do not promote or change aliases from a check request. Missing required health evidence means not yet verified.
+
+## Example requests
+
+- **Normal (inspect):** Check the candidate deployment before promotion; do not promote it.
+- **edge (inspect):** Check readiness for a deployment that removes a database column.
+- **blocked (inspect):** Assess readiness without current health evidence; do not promote aliases.

package/plugins/just-vibe/skills/vercel-routing/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: vercel-routing
+description: "Diagnose redirects, rewrites, headers, domains, and route behavior Use for platform/framework route interaction; vite-assets diagnoses emitted asset paths."
+---
+
+# vercel-routing
+
+Diagnose redirects, rewrites, headers, domains, and route behavior
+
+## Choose this workflow
+
+Use for platform/framework route interaction; vite-assets diagnoses emitted asset paths.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Vercel methods](../../references/packs/vercel.md) for tool selection and operational details. Resolve these paths from this skill file; all runtime assets ship inside the plugin.
+
+## Input and mode
+
+Use the complete request appended to this invocation, preserving all constraints and references. Default mode: **inspect**. Inspect; URL/path, expected response, project, and deployment configuration.
+
+exact team/project/environment and deployment/revision when applicable; read access to relevant configuration/logs. Verify installed CLI/API support and framework behavior during implementation. Never print environment values or infer promotion authorization from a preview request.
+
+Declared evidence requirements: `vercel.context`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Redirects, rewrites, headers, domains, and framework routing interactions.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Inspect precedence and base paths, trace bounded read-only requests, compare preview/production assumptions, and identify loops or unmatched paths.
+- Build a request table with host, path, method and expected handler; trace redirects, rewrites, filesystem and framework routing in effective order.
+
+## Decision branches
+
+- **When a catch-all rewrite returns HTML for API or asset paths:** Narrow the rule and verify reserved paths plus direct nested navigation.
+
+## Deliver and verify
+
+- Route trace, cause, and proposed configuration patch; apply on an explicit fix request.
+- Route matrix, observed status/location/content type and responsible rule.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Redirect loops are identified; API routes are not swallowed by a catch-all SPA rewrite.
+
+## Stop and recover
+
+- No DNS or custom-domain mutation implicitly. Distinguish framework behavior from platform configuration and unavailable network evidence.
+
+## Example requests
+
+- **Normal (inspect):** Explain why nested API routes receive the SPA page instead of JSON.
+- **edge (inspect):** Fix preview deep links without swallowing API requests.
+- **blocked (inspect):** Inspect routing configuration without DNS access or live request evidence.