just-vibe 0.7.0

package/plugins/just-vibe/skills/orient/SKILL.md

@@ -0,0 +1,58 @@
+---
+name: orient
+description: "Identify stack, structure, conventions, and actual working commands Use for first contact with an unfamiliar repository; use map for dependency detail and explain for one symbol."
+---
+
+# orient
+
+Identify stack, structure, conventions, and actual working commands
+
+## Choose this workflow
+
+Use for first contact with an unfamiliar repository; use map for dependency detail and explain for one symbol.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [General methods](../../references/packs/general.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; repository or directory plus optional onboarding goal. Requires readable project files; runtime checks are optional.
+
+Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Stack, entry points, package boundaries, instructions, and documented commands; no setup or dependency installation.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+1. Run the bundled inspector with the selected project root. Read relevant ancestor and discovered project instructions, manifests, scripts, and representative entry points.
+2. Explain how the requested feature fits the actual project, list commands with their package directories, and distinguish discovered commands from checks you actually ran.
+
+Task-specific method: Inspect manifests and scripts, sample representative modules, reconcile documentation with code, and identify the shortest path to the user's goal. Label unexecuted commands as inferred. Read manifests and actual entry points before listing scripts; follow one reachable user path and identify generated or vendored boundaries.
+
+## Decision branches
+
+- **When multiple packages or stale documentation:** Identify each package's role and distinguish configured commands from commands actually exercised.
+
+## Deliver and verify
+
+- Project orientation with file links, likely run/check commands, conventions, and unknowns.
+- Entry-point table: path, role, caller, verification command, unresolved assumption.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A monorepo yields package-specific commands; stale README instructions are identified rather than repeated as working.
+
+## Stop and recover
+
+- Unreadable roots or ambiguous projects produce a targeted blocker; never claim the application runs without evidence.
+
+## Example requests
+
+- **Normal (inspect):** Help me find how checkout works and which checks cover it.
+- **edge (inspect):** Orient this monorepo; the README describes an app that was removed.
+- **blocked (inspect):** Inspect this source archive without Git history or executing scripts.

package/plugins/just-vibe/skills/perf/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: perf
+description: "Measure a performance problem and improve its cause Use for measured performance repair; architecture scaling work needs workload-level evidence."
+---
+
+# perf
+
+Measure a performance problem and improve its cause
+
+## Choose this workflow
+
+Use for measured performance repair; architecture scaling work needs workload-level evidence.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [General methods](../../references/packs/general.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; slow operation, workload, performance objective, and measurement budget.
+
+Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Measured bottlenecks and focused optimization; no blanket rewrites or unbounded production load.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Establish repeatable baseline conditions, profile the path, target the dominant cause, and compare latency/resource use and correctness after the change.
+- Define a repeatable workload and metric, preserve baseline conditions, and isolate the dominant resource or wait before changing code.
+
+## Decision branches
+
+- **When a faster implementation changes output quality or consistency:** Reject the comparison or disclose the product tradeoff for an explicit decision.
+
+## Deliver and verify
+
+- Patch, measurement method, before/after results, and tradeoffs.
+- Workload, sample size, before/after measurements, correctness checks and tradeoffs.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Improvement persists under comparable workloads; optimized behavior remains correct at a relevant boundary.
+
+## Stop and recover
+
+- If measurement is unavailable, provide hypotheses without numerical claims. Stop when noise or resource limits prevent trustworthy conclusions.
+
+## Example requests
+
+- **Normal (apply):** Measure and fix slow product filtering; preserve search results.
+- **edge (apply):** Optimize a slow search without changing ranking or using a warmer cache baseline.
+- **blocked (inspect):** Inspect source without profiler data; provide hypotheses and a bounded measurement plan.

package/plugins/just-vibe/skills/plan/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: plan
+description: "Inspect the project and produce a concrete implementation plan Use for implementation sequencing against existing code; tasks breaks an accepted plan into work items."
+---
+
+# plan
+
+Inspect the project and produce a concrete implementation plan
+
+## Choose this workflow
+
+Use for implementation sequencing against existing code; tasks breaks an accepted plan into work items.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [General methods](../../references/packs/general.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; objective or existing spec plus constraints and scope. Requires repository inspection.
+
+Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+An executable implementation sequence grounded in this project; no feature implementation.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Find affected modules, check existing patterns, order dependencies, identify verification and rollout needs, and separate discovery tasks from known changes.
+- Connect each step to actual files, interfaces and a completion check; put discovery before changes that depend on uncertain contracts.
+
+## Decision branches
+
+- **When a required integration cannot be inspected:** Plan a contract seam and isolated fixture, then identify the live verification prerequisite separately.
+
+## Deliver and verify
+
+- File/component-level steps, success criteria, risks, recovery approach, and dependencies.
+- Ordered change table with target files, dependencies, checks and rollback boundaries.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A developer can start the first task without guessing its target; a missing integration is an explicit prerequisite rather than assumed available.
+
+## Stop and recover
+
+- Do not estimate unknown work as certain. Ask only for decisions that change the plan materially.
+
+## Example requests
+
+- **Normal (plan):** Plan adding saved filters to the existing search page without new dependencies.
+- **edge (plan):** Plan a backward-compatible API change with old clients still active.
+- **blocked (inspect):** Plan from this partial repository; external service schemas are unavailable.

package/plugins/just-vibe/skills/polish/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: polish
+description: "Refine hierarchy, spacing, typography, and interaction details Use for focused refinement of existing UI; design handles a larger visual direction change."
+---
+
+# polish
+
+Refine hierarchy, spacing, typography, and interaction details
+
+## Choose this workflow
+
+Use for focused refinement of existing UI; design handles a larger visual direction change.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [General methods](../../references/packs/general.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; existing UI, target screens, and desired refinements. Requires runnable UI or supplied render evidence.
+
+Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Typography, spacing, hierarchy, consistency, and interaction detail; preserve product structure and behavior.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Inspect baseline renders, prioritize visible issues, adjust shared tokens/components where appropriate, and compare the same states after changes.
+- Capture the same viewport/state before editing; fix the dominant hierarchy, spacing or alignment issue at the shared primitive when appropriate.
+
+## Decision branches
+
+- **When a visual change alters hit areas or focus visibility:** Recheck interaction and keyboard behavior before accepting the visual improvement.
+
+## Deliver and verify
+
+- Focused UI improvements and before/after evidence.
+- Comparable before/after views and the behavior preserved by the refinement.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Refined alignment or hierarchy is visible; keyboard focus and responsive behavior are preserved.
+
+## Stop and recover
+
+- Do not turn a polish task into a redesign. If rendering is unavailable, distinguish code changes from visually verified results.
+
+## Example requests
+
+- **Normal (apply):** Improve spacing and hierarchy on billing without changing the flow.
+- **edge (apply):** Polish a form without shrinking touch targets or losing error messages.
+- **blocked (inspect):** Review screenshots only; identify refinements without claiming rendered verification.

package/plugins/just-vibe/skills/pr/SKILL.md

@@ -0,0 +1,58 @@
+---
+name: pr
+description: "Prepare a focused diff summary, PR description, and validation notes Use to prepare a local PR title/body; github-pr handles remote PR identity and creation."
+---
+
+# pr
+
+Prepare a focused diff summary, PR description, and validation notes
+
+## Choose this workflow
+
+Use to prepare a local PR title/body; github-pr handles remote PR identity and creation.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [General methods](../../references/packs/general.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/base, intended change, and available verification results.
+
+Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Prepare a PR title/body and diff summary; submission is handled by `github-pr` when requested.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Inspect the actual diff, separate unrelated edits, explain changed behavior, and include only checks supported by evidence.
+- Base the description on the final diff and verified checks, account for generated artifacts, and state user-visible behavior before implementation details.
+- All changes are owned by the user. Add no agent/model self-attribution, AI-generated signature, badge, or agent Co-authored-by trailer to commits, PRs, comments, release notes or messages. Use the existing user Git identity; preserve legitimate human attribution and required third-party notices.
+
+## Decision branches
+
+- **When local changes are absent from the pushed head:** Identify them as pending and do not describe them as submitted.
+
+## Deliver and verify
+
+- Reviewable title, description, risk notes, and readiness gaps.
+- Title, problem/behavior description, validation and readiness gaps.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Description matches the final diff; unrun tests are not listed as passing.
+- Review newly prepared commit/PR/message text, including template or hook additions, for agent self-attribution before submission; verify the resulting artifact when available. Do not silently rewrite existing history or remove human credits.
+
+## Stop and recover
+
+- Flag unresolved conflicts, missing base, or unintended files. Do not push, open a PR, or message reviewers from a preparation-only request.
+
+## Example requests
+
+- **Normal (plan):** Draft a PR title and description for this diff; do not submit it.
+- **edge (plan):** Prepare a PR description after the implementation scope changed.
+- **blocked (inspect):** Draft from the local diff without remote access or posting anything.

package/plugins/just-vibe/skills/profile/SKILL.md

@@ -0,0 +1,66 @@
+---
+name: profile
+description: "Set, inspect, automatically select or clear task-scoped engineering profiles Use when the user wants to change the agent’s working priorities or asks it to choose an appropriate role; this does not execute a product task by itself."
+---
+
+# profile
+
+Set, inspect, automatically select or clear task-scoped engineering profiles
+
+## Choose this workflow
+
+Use when the user wants to change the agent’s working priorities or asks it to choose an appropriate role; this does not execute a product task by itself.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [General methods](../../references/packs/general.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-read-only; changing in-context task selection does not grant apply mode. Preserve all appended task constraints.
+
+Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Relevant catalog entries, current task instructions and enough project evidence to choose responsibly.
+
+Only active task context or a supplied run record returned as JSON. Persist to a user-requested location only when explicitly asked; never change global host settings.
+
+## Execute
+
+1. Read [profile selection](../../references/profiles.md), then use toolkit profiles and toolkit profile ID to find and inspect relevant roles.
+2. Keep simple selection in task context. With an existing run, call session profile using run and selection; with a new explicit single role, workflow COMMAND --profile ID preserves it. Do not create a fresh run just to evade a prior pin or budget.
+3. State the selected profile briefly once. Apply relevant priorities in subsequent workflows without repeating role claims or forcing every suggested workflow to run.
+
+Task-specific method: Choose one primary role and at most two distinct complementary roles; read only their role references. Use task evidence, not a file extension alone. Explicit user selections are pinned for the current task by default. Agent selections are unpinned and cannot replace or clear a pinned user selection. For auto, an explicit user request first clears the pin, then the agent selects with a concrete reason. Clear removes role emphasis; status reports the current selection without changing it. A profile in project preferences is a candidate, not a pin or permission. Apply it only when it fits the current request and does not conflict with the user’s task selection.
+
+## Read when relevant
+
+- Saving requested preferences, decisions or a named continuation: [Project continuity](../../references/daily-workflows.md).
+
+## Decision branches
+
+- **When the user requests auto after pinning a role:** Record a user clear with the actual request as reason, then record an unpinned agent selection based on task evidence.
+- **When the agent discovers a new focus while a user selection is pinned:** Retain the selection and apply required task checks without silently changing roles.
+
+## Deliver and verify
+
+- Primary and secondary role IDs, selection source, task scope, pin state and short evidence-based reason.
+- Relevant priorities, checks and conflicts with the task brief; original scope and permissions stay intact.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Pinned user choices survive automatic routing. Profile changes preserve brief, mode, scope, authorization, budgets, attempts and original success criteria.
+- A profile never creates a subagent, installs tools, changes credentials or claims professional expertise.
+
+## Stop and recover
+
+- Keep an existing pin when an inferred role conflicts; ask only if an explicit conflicting request needs resolution.
+- If no role fits, continue with ordinary task guidance and state that limitation. Do not invent a shipped role.
+
+## Example requests
+
+- **normal (inspect):** Set machine-learning-engineer for this implementation, with mlops-engineer as a secondary focus.
+- **edge (inspect):** Use principal-engineer to review this design, but keep the change local and do not redesign the platform.
+- **blocked (inspect):** Choose automatically while frontend-engineer is pinned; preserve the pin and explain relevant task checks.

package/plugins/just-vibe/skills/profiles/SKILL.md

@@ -0,0 +1,58 @@
+---
+name: profiles
+description: "Browse engineering role profiles and compare their priorities, boundaries and verification Use to discover roles; profile selects or clears the active task role, while tools discovers actions."
+---
+
+# profiles
+
+Browse engineering role profiles and compare their priorities, boundaries and verification
+
+## Choose this workflow
+
+Use to discover roles; profile selects or clears the active task role, while tools discovers actions.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [General methods](../../references/packs/general.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 only. Accept a role, discipline, level or task description as appended context.
+
+Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+The shipped profile catalog and supplied task context.
+
+No changes to active profile, project files, global settings or external state.
+
+## Execute
+
+1. Run toolkit profiles with the full relevant query; for specific details run toolkit profile ID. Read [profile selection](../../references/profiles.md) for scope and precedence.
+2. Return a focused comparison and useful examples. Search is lexical discovery, not an authority or competence score.
+
+Task-specific method: Search the profile catalog and group relevant roles by discipline; separate architecture specialties from seniority-oriented profiles. Explain the decisive difference between nearby roles using their priorities, decision branch and verification; do not equate titles with capabilities or seniority credentials.
+
+## Decision branches
+
+- **When several profiles appear relevant:** Compare primary responsibility and actual task evidence; show options without silently activating one.
+
+## Deliver and verify
+
+- Relevant role IDs, purpose, priorities and example use.
+- Clear distinctions between close profiles and their suggested workflows.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Every listed profile exists in the installed catalog. Discovery leaves the active task selection unchanged.
+
+## Stop and recover
+
+- Do not activate a role, install tools or execute suggested workflows from a discovery request.
+
+## Example requests
+
+- **normal (inspect):** Show profiles for machine learning engineering and model operations.
+- **edge (inspect):** Compare senior, staff, principal and software architect profiles for this migration.
+- **blocked (inspect):** Find a profile for an unsupported specialty without pretending it is installed.

package/plugins/just-vibe/skills/react-async/SKILL.md

@@ -0,0 +1,57 @@
+---
+name: react-async
+description: "Fix loading races, cancellation, stale responses, and async behavior Use for request races, optimistic updates and async states; react-effects handles general lifecycle synchronization."
+---
+
+# react-async
+
+Fix loading races, cancellation, stale responses, and async behavior
+
+## Choose this workflow
+
+Use for request races, optimistic updates and async states; react-effects handles general lifecycle synchronization.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [React methods](../../references/packs/react.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; async interaction, request/cache layer, and observed race or loading issue.
+
+component source, React/framework versions, state/data conventions, and relevant test tooling. Browser/profiler evidence is needed for measured rendering claims. Preserve existing framework and state libraries unless changing them is part of the request.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Cancellation, response ownership, loading/error transitions, and optimistic recovery.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Trace resource identity, the owner of each request, component lifetime and every state publication path: loading, success, error and optimistic reconciliation. Define which completion is current after navigation, account changes or a new selection.
+- Use the existing framework/data layer mechanism to separate stale-result suppression from actual cancellation. Shared requests may outlive one component; cancelling one subscriber must not invalidate another subscriber’s result.
+- Guard both success and failure publication against stale identity and disposal. Clean up subscriptions/listeners on all terminal paths and prevent disposed owners from starting further work unless the lifecycle contract explicitly permits reactivation.
+- Control completion order in tests: newer success before older success, newer success before older failure, unmount while pending and shared-request cancellation. For optimistic writes, reconcile from authoritative state after ambiguous completion instead of assuming abort undid the server effect.
+
+## Decision branches
+
+- **When cancellation arrives after the server applied a write:** Reconcile authoritative state rather than assuming the business effect was undone.
+
+## Deliver and verify
+
+- Request ownership and state transitions, repair, deterministic stale-success/stale-error/disposal checks and optimistic reconciliation limits.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Neither an older success nor an older error can replace current state. Disposal releases owned resources and prevents later publication; shared consumers remain independent.
+
+## Stop and recover
+
+- Do not assume cancellation undoes server effects. Preserve the existing fetching library unless replacement is explicitly justified in scope.
+
+## Example requests
+
+- **Normal (apply):** Fix out-of-order search responses overwriting newer results.
+- **edge (apply):** Fix search results that revert when older requests finish last.
+- **blocked (inspect):** Inspect async behavior without live network access; use a controlled deferred-response fixture.

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

@@ -0,0 +1,56 @@
+---
+name: react-audit
+description: "Review components, hooks, state ownership, and behavioral risks Use for React correctness inspection; react-rerenders requires performance evidence and react-effects targets synchronization."
+---
+
+# react-audit
+
+Review components, hooks, state ownership, and behavioral risks
+
+## Choose this workflow
+
+Use for React correctness inspection; react-rerenders requires performance evidence and react-effects targets synchronization.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [React methods](../../references/packs/react.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; component tree/feature and behavioral concerns.
+
+component source, React/framework versions, state/data conventions, and relevant test tooling. Browser/profiler evidence is needed for measured rendering claims. Preserve existing framework and state libraries unless changing them is part of the request.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Hooks, ownership, identity, effects, async boundaries, and component contracts.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Trace props/state and lifecycle, identify reachable failure paths, inspect tests and framework constraints, and rank concrete risks.
+- Identify React/framework version and server/client boundary, then trace state ownership, hook order, key identity and reachable interaction states.
+
+## Decision branches
+
+- **When source suggests excess rendering without a profile:** Report a performance hypothesis separately from demonstrated correctness defects.
+
+## Deliver and verify
+
+- Findings with component locations, triggers, and repair suggestions.
+- Finding location, triggering interaction, state transition and severity rationale.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A conditional hook or unstable key problem includes its trigger; a stylistic preference is not presented as a correctness bug.
+
+## Stop and recover
+
+- Do not rewrite components during an audit or infer runtime rendering cost solely from source size.
+
+## Example requests
+
+- **Normal (inspect):** Audit checkout hooks and state ownership for reachable behavioral bugs.
+- **edge (inspect):** Audit a reordered editable list and a component with an early return before a hook.
+- **blocked (inspect):** Review React source without a browser; mark interaction behavior unverified.

package/plugins/just-vibe/skills/react-component/SKILL.md

@@ -0,0 +1,65 @@
+---
+name: react-component
+description: "Build a component with its states, API, and accessibility Use to implement one component with a clear contract; ui-system defines shared primitives and tokens."
+---
+
+# react-component
+
+Build a component with its states, API, and accessibility
+
+## Choose this workflow
+
+Use to implement one component with a clear contract; ui-system defines shared primitives and tokens.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [React methods](../../references/packs/react.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; purpose, props/API, visual constraints, states, and usage context.
+
+component source, React/framework versions, state/data conventions, and relevant test tooling. Browser/profiler evidence is needed for measured rendering claims. Preserve existing framework and state libraries unless changing them is part of the request.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Reusable component plus necessary integration/example; no unrelated design-system replacement.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Inspect existing primitives and the consumer contract. Choose state ownership, semantics and interaction behavior before implementing the narrow component API.
+- Select only the relevant scenario guide for dialogs, comboboxes or date inputs; implement normal and recovery states, then verify real interactions and parent-controlled updates.
+
+## Read when relevant
+
+- Implementing a dialog or modal overlay: [dialog interaction](../../references/scenarios/dialog.md).
+- Implementing selection or autocomplete: [combobox interaction](../../references/scenarios/combobox.md).
+- Implementing a date or range input: [date-picker interaction](../../references/scenarios/date-picker.md).
+
+## Decision branches
+
+- **When the component opens an overlay:** Specify initial/contained/return focus and dismissal with the dialog guide.
+- **When search text and selection are different states:** Use the combobox guide for ownership, keyboard/IME behavior and late search results.
+- **When the value contains dates or ranges:** Resolve date-only versus instant semantics, locale, invalid input and range boundaries before choosing storage.
+- **When parent state changes after mount:** Preserve a consistent controlled ownership contract and test reset and two independent instances.
+
+## Deliver and verify
+
+- Component, documented contract, usage, and meaningful checks.
+- Component API, rendered states, keyboard behavior and consumer example.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Controlled updates behave consistently; keyboard users can perform the primary interaction.
+
+## Stop and recover
+
+- Avoid speculative configuration options and new libraries without need. Missing design details use established conventions rather than blocking routine work.
+
+## Example requests
+
+- **Normal (apply):** Build an accessible controlled date-range component using existing primitives.
+- **edge (apply):** Build a reusable selector supporting controlled updates and an empty option list.
+- **blocked (inspect):** Specify a component using existing primitives when design assets are unavailable.

package/plugins/just-vibe/skills/react-effects/SKILL.md

@@ -0,0 +1,58 @@
+---
+name: react-effects
+description: "Investigate effect loops, stale closures, races, and missing cleanup Use for synchronization, cleanup or dependency defects; react-state handles authoritative data placement."
+---
+
+# react-effects
+
+Investigate effect loops, stale closures, races, and missing cleanup
+
+## Choose this workflow
+
+Use for synchronization, cleanup or dependency defects; react-state handles authoritative data placement.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [React methods](../../references/packs/react.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 for a reported effect bug; component, symptom, and expected synchronization.
+
+component source, React/framework versions, state/data conventions, and relevant test tooling. Browser/profiler evidence is needed for measured rendering claims. Preserve existing framework and state libraries unless changing them is part of the request.
+
+Declared evidence requirements: `project.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Effect dependencies, cleanup, stale closures, loops, and external synchronization.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Classify each effect as synchronization with an external system or a derived computation. Derive render-only values directly where appropriate; do not add state/effects merely to mirror existing props.
+- Trace dependency identity through setup, dependency change, cleanup and unmount. Check development replay/remount behavior against the installed framework version; cleanup must undo the resource acquired by that setup instance.
+- For async synchronization, protect current identity on both fulfillment and rejection and define ownership of any shared work. Avoid suppressing dependency checks or using a permanent once flag to hide an incorrect lifetime.
+- Verify rapid identity changes and repeated setup/cleanup with observable subscriptions, state and resource counts. Distinguish a verified lifecycle fix from a claimed performance improvement that has not been measured.
+
+## Decision branches
+
+- **When old async work can complete after a new selection:** Guard stale completion as well as cleaning up; cancellation alone does not establish which response is current.
+
+## Deliver and verify
+
+- Effect repair and lifecycle/regression evidence.
+- Effect purpose, dependency/lifecycle trace and remount/race regression checks.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Account changes do not show stale responses; repeated setup/cleanup does not leak subscriptions or duplicate effects.
+
+## Stop and recover
+
+- Do not silence dependency warnings to hide the issue. Preserve intended behavior across the framework's development checks.
+
+## Example requests
+
+- **Normal (apply):** Fix stale account data caused by effect request races.
+- **edge (apply):** Fix an account panel where a late response from the prior account overwrites the current one.
+- **blocked (inspect):** Audit effect source without reproducing browser timing; identify the required controlled race test.