just-vibe 0.7.0

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

@@ -0,0 +1,56 @@
+---
+name: debug
+description: "Investigate and explain a failure before changing code Use to identify a cause and next experiment; fix applies a requested repair."
+---
+
+# debug
+
+Investigate and explain a failure before changing code
+
+## Choose this workflow
+
+Use to identify a cause and next experiment; fix applies a requested repair.
+
+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; failure evidence, environment, and expected behavior. Running a reproduction that writes requires apply mode or explicit execution authorization.
+
+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
+
+Diagnosis and narrowing; no automatic code repair.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Build a hypothesis list, inspect logs and code, seek evidence that distinguishes causes, and identify the smallest next experiment.
+- Rank hypotheses by discriminating observations, trace the first divergence from expected behavior, and use bounded probes rather than repeated full runs.
+
+## Decision branches
+
+- **When logs establish symptoms but not causation:** Report competing hypotheses and the lowest-cost observation that separates them.
+
+## Deliver and verify
+
+- Most supported cause, supporting/contradicting evidence, and a proposed fix or reproduction step.
+- Hypothesis table with supporting/contradicting evidence and the next targeted probe.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Eliminates an initially plausible false cause; insufficient evidence yields ranked hypotheses, not false certainty.
+
+## Stop and recover
+
+- Avoid repeated identical probes. Do not mutate production or install debugging tools implicitly.
+
+## Example requests
+
+- **Normal (inspect):** Investigate intermittent checkout failures from these logs; do not change files.
+- **edge (inspect):** Diagnose a timeout that occurs only after a successful database write.
+- **blocked (inspect):** Diagnose using redacted logs only; do not restart services or infer missing spans.

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

@@ -0,0 +1,57 @@
+---
+name: decide
+description: "Recommend an option against explicit requirements Use to recommend one bounded choice; brainstorm expands options and compare describes differences."
+---
+
+# decide
+
+Recommend an option against explicit requirements
+
+## Choose this workflow
+
+Use to recommend one bounded choice; brainstorm expands options and compare describes differences.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Decisions methods](../../references/packs/decisions.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; decision, options, hard requirements, and priorities.
+
+the decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Recommend one bounded choice; no purchase, installation, or commitment.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Write the decision question, outcome, deadline, hard constraints and current/default option. Eliminate confirmed infeasible choices before scoring preferences; unknown feasibility is pending evidence, not a middle score.
+- Compare remaining options using evidence with dates and scope. Separate observed facts, estimates and preferences; avoid double-counting correlated criteria or presenting subjective weights as objective truth.
+- Identify the assumption capable of changing the recommendation. Vary plausible scores/weights or bounds and report whether the preferred option changes; use a bounded discriminating spike when its information is worth the delay.
+- Recommend an option or conditional reversible choice, explain consequences and rejected alternatives, and set an observable revisit trigger. Record adoption only when it occurred; a recommendation does not commit stakeholders or authorize a purchase.
+
+## Decision branches
+
+- **When evidence cannot distinguish feasible options:** Make a reversible conditional choice or propose a discriminating spike rather than fake certainty.
+
+## Deliver and verify
+
+- Decision and recommendation, feasibility/evidence table, decisive uncertainty or sensitivity, consequences, adoption status and revisit trigger.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A hard requirement outweighs a cosmetic advantage; insufficient evidence produces a conditional recommendation.
+
+## Stop and recover
+
+- Ask only for preferences that can flip the outcome. Do not fabricate stakeholder agreement or imply the recommendation is already adopted.
+
+## Example requests
+
+- **Normal (plan):** Choose a job queue approach given our small team and existing Postgres service.
+- **edge (plan):** Choose a cache strategy when consistency is mandatory but traffic is uncertain.
+- **blocked (inspect):** Recommend a conditional choice with no stakeholder cost weights.

package/plugins/just-vibe/skills/decision-adr/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: decision-adr
+description: "Write a decision record with alternatives and consequences Use to record a proposed or adopted architectural decision; decide determines a recommendation first."
+---
+
+# decision-adr
+
+Write a decision record with alternatives and consequences
+
+## Choose this workflow
+
+Use to record a proposed or adopted architectural decision; decide determines a recommendation first.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Decisions methods](../../references/packs/decisions.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; proposed/adopted decision, context, alternatives, and destination when saving.
+
+the decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Record one architectural decision with status and consequences; no new approval or implementation.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Inspect existing ADR conventions, separate actual agreement from recommendation, capture rationale and tradeoffs, and link superseded decisions.
+- Follow existing numbering/status conventions; include context, alternatives, consequences and links to superseded records without rewriting history.
+
+## Decision branches
+
+- **When adoption is not confirmed:** Keep status proposed and distinguish the recommendation from actual agreement.
+
+## Deliver and verify
+
+- ADR with context, decision, alternatives, consequences, status, and revisit triggers.
+- ADR text with status, rationale, consequences and supersession links where applicable.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A proposed choice remains proposed; a superseding ADR preserves the older record and links it.
+
+## Stop and recover
+
+- Do not manufacture dates, participants, or consent. Save only in the requested or established documentation location.
+
+## Example requests
+
+- **Normal (plan):** Draft an ADR for the proposed shared identity service; keep status proposed.
+- **edge (plan):** Write an ADR that supersedes an earlier database decision.
+- **blocked (inspect):** Draft an ADR without known meeting dates or approvers; do not invent them.

package/plugins/just-vibe/skills/decision-buy-build/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: decision-buy-build
+description: "Compare building, buying, and integrating a solution Use for vendor versus internal capability decisions; research verifies current vendor claims."
+---
+
+# decision-buy-build
+
+Compare building, buying, and integrating a solution
+
+## Choose this workflow
+
+Use for vendor versus internal capability decisions; research verifies current vendor claims.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Decisions methods](../../references/packs/decisions.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; capability, team capacity, compliance/integration constraints, volume, and budget.
+
+the decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Build, buy, and hybrid options across initial and ongoing ownership.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Compare fit, integration, maintenance, migration, service dependence, and total-cost assumptions; verify current vendor capabilities when relevant.
+- Compare integration, operations, staffing, exit/export and failure ownership over a stated usage horizon; include the current workaround.
+
+## Decision branches
+
+- **When a vendor lacks a hard requirement or usable export:** Exclude it or state the explicit compromise before calculating weighted convenience.
+
+## Deliver and verify
+
+- Option comparison, recommendation, cost model inputs, and a validation/exit plan.
+- Fit/gap matrix, cost assumptions, ownership burden and exit plan.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Maintenance and migration effort appear alongside subscription cost; a vendor missing a hard feature is excluded.
+
+## Stop and recover
+
+- No purchases or account creation. Do not convert speculative usage into precise budget claims.
+
+## Example requests
+
+- **Normal (plan):** Compare building and buying organization authentication with future SSO.
+- **edge (plan):** Compare buying search with building it when private indexing is required.
+- **blocked (inspect):** Assess buy versus build with no price quote; keep uncertain costs as ranges.

package/plugins/just-vibe/skills/decision-matrix/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: decision-matrix
+description: "Compare options using weighted criteria and explain the weights Use when several explicit criteria need weighted comparison; decide suffices for a decisive hard constraint."
+---
+
+# decision-matrix
+
+Compare options using weighted criteria and explain the weights
+
+## Choose this workflow
+
+Use when several explicit criteria need weighted comparison; decide suffices for a decisive hard constraint.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Decisions methods](../../references/packs/decisions.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; options, measurable criteria, weights or priority guidance.
+
+the decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Transparent weighted comparison, including hard exclusions and sensitivity analysis.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Define score scales, cite evidence for scores, separate must-haves, normalize weights, and vary uncertain values to test ranking stability.
+- Define score anchors before rating, keep feasibility separate, and vary uncertain weights/scores over plausible ranges.
+
+## Decision branches
+
+- **When a small plausible perturbation changes the winner:** Report the unstable ranking and seek evidence on the sensitive criterion.
+
+## Deliver and verify
+
+- Matrix with score rationale, assumptions, sensitivity, and recommendation.
+- Criterion definitions, sources, exclusions, weighted calculation and sensitivity result.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- An infeasible option cannot win by averaging; small weight changes that flip the result are disclosed.
+
+## Stop and recover
+
+- Unknown scores remain unknown or bounded. Avoid false precision and do not invent numeric confidence from subjective ratings.
+
+## Example requests
+
+- **Normal (plan):** Compare three authentication options; maintenance is the highest priority.
+- **edge (plan):** Compare three databases when one fails a hard regional requirement.
+- **blocked (inspect):** Build a comparison with unknown costs; use bounds or missing values instead of invented scores.

package/plugins/just-vibe/skills/decision-premortem/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: decision-premortem
+description: "Assume a proposal failed and identify plausible causes Use to analyze plausible future failure of a proposal; ops-postmortem reconstructs an actual incident."
+---
+
+# decision-premortem
+
+Assume a proposal failed and identify plausible causes
+
+## Choose this workflow
+
+Use to analyze plausible future failure of a proposal; ops-postmortem reconstructs an actual incident.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Decisions methods](../../references/packs/decisions.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; proposal, success definition, operating context, and time horizon.
+
+the decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Plausible reasons the proposal could fail and preventive actions.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Assume the outcome failed, trace realistic causal chains, rank impact/likelihood, identify early signals, and propose proportionate mitigations.
+- Work backward from a concrete failed outcome through design choices, contributing conditions and observable warning signs.
+
+## Decision branches
+
+- **When a risk cannot be connected to this proposal:** Remove the generic warning and focus on mechanisms supported by context.
+
+## Deliver and verify
+
+- Failure scenarios, warning indicators, mitigations, and untested assumptions.
+- Failure chain, early signal, mitigation, response and residual uncertainty.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Risks connect to specific proposal choices; a mitigation includes a detectable signal or concrete action.
+
+## Stop and recover
+
+- Do not present hypothetical failures as observed incidents or pad the report with generic catastrophes unrelated to the design.
+
+## Example requests
+
+- **Normal (plan):** Identify plausible ways the migration plan could fail and early warning signs.
+- **edge (plan):** Premortem a rollout whose rollback cannot undo generated data.
+- **blocked (inspect):** Analyze hypothetical failure without treating it as an observed incident.

package/plugins/just-vibe/skills/decision-reversible/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: decision-reversible
+description: "Separate reversible choices from expensive commitments Use to classify and reduce switching cost; decision-spike tests a specific uncertainty."
+---
+
+# decision-reversible
+
+Separate reversible choices from expensive commitments
+
+## Choose this workflow
+
+Use to classify and reduce switching cost; decision-spike tests a specific uncertainty.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Decisions methods](../../references/packs/decisions.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; candidate decisions, migration constraints, contracts, and timeline.
+
+the decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Reversibility, lock-in, exit cost, and staged commitment.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Identify persisted data, external promises, switching mechanisms, and operational costs; classify reversibility with evidence and design cheaper experiments.
+- Inventory persisted data, external promises and dependency adoption; distinguish rollback of code from reversal of consequences.
+
+## Decision branches
+
+- **When a configuration toggle has irreversible downstream effects:** Classify those effects separately and design a bounded trial with cleanup or reconciliation.
+
+## Deliver and verify
+
+- Decision map with reversal steps, dependencies, cost drivers, and commitment checkpoints.
+- Reversibility dimensions, exit cost assumptions and experiment boundary.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A simple config toggle with irreversible downstream data effects is not marked freely reversible; a trial has a usable exit path.
+
+## Stop and recover
+
+- Avoid unsupported effort estimates. Unknown contractual constraints remain explicit dependencies.
+
+## Example requests
+
+- **Normal (plan):** Identify irreversible commitments in this storage-provider decision.
+- **edge (plan):** Assess a feature flag that sends customer emails.
+- **blocked (inspect):** Assess reversibility with unknown contract terms; identify the missing constraint.

package/plugins/just-vibe/skills/decision-revisit/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: decision-revisit
+description: "Reassess a decision using new constraints or evidence Use when new evidence may invalidate a recorded choice; decision-adr preserves the resulting decision record."
+---
+
+# decision-revisit
+
+Reassess a decision using new constraints or evidence
+
+## Choose this workflow
+
+Use when new evidence may invalidate a recorded choice; decision-adr preserves the resulting decision record.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Decisions methods](../../references/packs/decisions.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; prior decision record, new evidence, and changed constraints.
+
+the decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Reassess a specific choice while preserving its historical rationale.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Identify original assumptions, compare them to current facts, estimate transition consequences, and recommend retaining, modifying, or superseding the decision.
+- Compare original assumptions with current facts and include migration costs, disruption and option value in retaining versus replacing.
+
+## Decision branches
+
+- **When only preferences or fashion changed:** Retain the decision unless an explicit new priority justifies transition costs.
+
+## Deliver and verify
+
+- Reassessment with changed premises, alternatives, migration implications, and proposed status update.
+- Changed assumption table, retain/change recommendation and adoption/revisit conditions.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- New fashion alone does not invalidate a sound decision; a broken core assumption triggers a concrete alternative analysis.
+
+## Stop and recover
+
+- Do not rewrite history or silently mark a decision superseded without an adoption instruction.
+
+## Example requests
+
+- **Normal (plan):** Reassess the cache decision now that writes occur in two regions.
+- **edge (plan):** Revisit a queue choice after workload ordering requirements change.
+- **blocked (inspect):** Review an old decision without its rationale; distinguish reconstruction from recorded facts.

package/plugins/just-vibe/skills/decision-spike/SKILL.md

@@ -0,0 +1,58 @@
+---
+name: decision-spike
+description: "Design or run a bounded experiment to resolve uncertainty Use for a bounded experiment resolving a decision; build creates production behavior."
+---
+
+# decision-spike
+
+Design or run a bounded experiment to resolve uncertainty
+
+## Choose this workflow
+
+Use for a bounded experiment resolving a decision; build creates production behavior.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Decisions methods](../../references/packs/decisions.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; uncertainty, competing hypotheses, success criteria, time/compute budget, and scratch scope.
+
+the decision question, constraints, alternatives or permission to identify them, and relevant project evidence. Current vendor claims and prices require current authoritative sources during execution. Scores are decision aids, not facts.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Design or explicitly run a disposable experiment; no production integration disguised as a spike.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Name the decision the spike will inform and the uncertainty that could change it. Define a representative workload/sample and an observable acceptance/rejection criterion before building.
+- Set a time/compute/data budget, isolated artifact location and stop rule. Choose the smallest prototype or measurement that can distinguish the alternatives; avoid building production scaffolding that does not test the hypothesis.
+- Execute within authorization and record raw observations, versions and deviations from the plan. Preserve failures and resource-limit outcomes; do not change the success threshold after seeing the result.
+- Classify supported, rejected or inconclusive. Explain the implication for the original decision and the limits of the sample; prototype success does not establish production readiness.
+
+## Decision branches
+
+- **When budget expires without separation:** Preserve measurements and report inconclusive, including what a larger experiment would need.
+
+## Deliver and verify
+
+- Experiment protocol or results, reusable evidence, and decision implications.
+- Hypotheses, experiment, resource cap, observations and decision implication.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- The test can reject a hypothesis; reaching the budget yields an inconclusive result rather than endless expansion.
+
+## Stop and recover
+
+- Paid resources require budget authorization. Keep prototype quality and production readiness clearly distinct.
+
+## Example requests
+
+- **Normal (plan):** Design a two-hour local spike to test whether our current search index is sufficient.
+- **edge (plan):** Test whether streaming reduces perceived latency without building the full product.
+- **blocked (inspect):** Design a spike without available hardware; do not fabricate benchmark results.

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

@@ -0,0 +1,56 @@
+---
+name: deploy
+description: "Prepare or perform deployment within the requested authorization Use for an explicitly targeted deployment or its plan; vercel-preview is the Vercel preview specialization."
+---
+
+# deploy
+
+Prepare or perform deployment within the requested authorization
+
+## Choose this workflow
+
+Use for an explicitly targeted deployment or its plan; vercel-preview is the Vercel preview specialization.
+
+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; application, environment/account, artifact/revision, and desired deployment action.
+
+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 deployment; execute only the deployment explicitly requested and authorized, including promotion separately when needed.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Resolve target, inspect prerequisites and checks, identify recovery, execute authorized steps, and verify the deployed revision and health.
+- Confirm immutable artifact/revision, environment and health criteria; resolve schema compatibility and the previous usable target before execution.
+
+## Decision branches
+
+- **When submission times out with uncertain provider state:** Look up the operation by revision or deployment ID before creating another deployment.
+
+## Deliver and verify
+
+- Deployment plan or actual deployment ID/URL, health evidence, and recovery information.
+- Target, revision, deployment identity, health results and recoverability limits.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Successful upload without healthy startup is not completion; an ambiguous production target blocks execution.
+
+## Stop and recover
+
+- Do not create paid resources, alter DNS, or promote previews implicitly. Report partial deployment state before retrying.
+
+## Example requests
+
+- **Normal (plan):** Prepare a staging deployment plan; identify the exact revision and rollback.
+- **edge (plan):** Deploy a preview whose build succeeds but startup health fails.
+- **blocked (inspect):** Plan deployment without provider access; do not claim a URL or deployed revision.

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

@@ -0,0 +1,56 @@
+---
+name: deps
+description: "Assess dependency updates and compatibility Use for dependency assessment or requested updates; vite-upgrade handles Vite-specific migration behavior."
+---
+
+# deps
+
+Assess dependency updates and compatibility
+
+## Choose this workflow
+
+Use for dependency assessment or requested updates; vite-upgrade handles Vite-specific migration behavior.
+
+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; package scope, update goal, compatibility constraints, and registry access when needed.
+
+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
+
+Dependency health and upgrade proposals; requested updates select apply mode and include lockfiles.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Inspect manifests/resolution, check current release notes and advisories, identify peer/runtime constraints, and group changes by risk.
+- Read resolved versions and peer/runtime ranges; identify why each dependency exists and isolate direct changes from lockfile churn.
+
+## Decision branches
+
+- **When an advisory has configuration-dependent exposure:** Trace actual reachable use and report that condition before recommending a breaking upgrade.
+
+## Deliver and verify
+
+- Prioritized recommendations or authorized update with compatibility checks.
+- Current/target versions, compatibility risks, advisory evidence and update checks.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A peer conflict is detected; updating one package does not silently churn unrelated dependency versions.
+
+## Stop and recover
+
+- Never treat latest as automatically best. Missing registry information is unknown, not proof of no vulnerabilities.
+
+## Example requests
+
+- **Normal (inspect):** Assess a compatible dependency update without changing files yet.
+- **edge (inspect):** Update one dependency with a conflicting peer and an unrelated dirty lockfile.
+- **blocked (inspect):** Assess dependencies from the lockfile with unavailable advisory access.