just-vibe 0.7.0

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

@@ -0,0 +1,59 @@
+---
+name: checkpoint
+description: "Save progress, evidence, and unresolved work Use for a compact continuation snapshot; handoff adds context for a different reader."
+---
+
+# checkpoint
+
+Save progress, evidence, and unresolved work
+
+## Choose this workflow
+
+Use for a compact continuation snapshot; handoff adds context for a different reader.
+
+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; current task, destination if supplied, and continuity needs.
+
+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
+
+Save task state in a local approved artifact; no permanent behavioral memory or external posting.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Save the objective, constraints, decisions, completed evidence, remaining work and next step using project checkpoint NAME with the current revision when structured storage is appropriate. The helper captures repository/worktree identity; keep external operation IDs in the task context without credentials.
+
+## Read when relevant
+
+- Saving requested preferences, decisions or a named continuation: [Project continuity](../../references/daily-workflows.md).
+
+## Decision branches
+
+- **When a test predates intervening edits:** Retain its old revision and mark current verification pending.
+
+## Deliver and verify
+
+- Concise checkpoint with timestamp and evidence references.
+- Timestamped state, evidence references and one executable next action.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A later session can locate the work and distinguish done from pending; stale checks retain the revision they actually tested.
+
+## Stop and recover
+
+- Do not overwrite an unrelated checkpoint. Ask about destination only when no established project location can be inferred.
+
+## Example requests
+
+- **Normal (apply):** Save the current task state to the existing project checkpoint file.
+- **edge (apply):** Checkpoint work after a successful test followed by additional edits.
+- **blocked (inspect):** Summarize current work without saving files or implying unavailable checks passed.

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

@@ -0,0 +1,56 @@
+---
+name: ci
+description: "Diagnose or improve continuous integration Use for CI diagnosis across providers; github-fix-ci handles GitHub run identity and requested repairs."
+---
+
+# ci
+
+Diagnose or improve continuous integration
+
+## Choose this workflow
+
+Use for CI diagnosis across providers; github-fix-ci handles GitHub run identity and requested repairs.
+
+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 for diagnosis; apply for requested fixes. Requires workflow files and relevant runner logs.
+
+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
+
+Repository CI behavior; GitHub-specific operations use the GitHub pack.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Locate the first meaningful failure, compare runner/local environments, distinguish code failure from infrastructure, and validate authorized workflow changes.
+- Find the first causal failure in the job graph, compare effective runtime and lockfile inputs, and distinguish required failures from downstream cancellations.
+
+## Decision branches
+
+- **When logs show a provider outage rather than changed-code failure:** Preserve evidence and recommend a bounded rerun without weakening checks.
+
+## Deliver and verify
+
+- Root-cause evidence or focused patch, local checks, and remaining remote validation.
+- Run/revision identity, causal log excerpt, environment difference and remaining verification.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A cache-related failure is not hidden by ignoring errors; unrelated matrix jobs remain intact.
+
+## Stop and recover
+
+- Do not rerun costly jobs or change organization secrets without the relevant request. Report inaccessible logs explicitly.
+
+## Example requests
+
+- **Normal (inspect):** Diagnose the failed matrix job from these logs without rerunning it.
+- **edge (inspect):** Diagnose one failing matrix job without disabling the other jobs.
+- **blocked (inspect):** Inspect supplied CI logs without runner access or triggering a rerun.

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

@@ -0,0 +1,56 @@
+---
+name: cleanup
+description: "Remove verified dead code and unnecessary complexity Use for evidence-backed removal of unnecessary code or assets; refactor preserves behavior through restructuring."
+---
+
+# cleanup
+
+Remove verified dead code and unnecessary complexity
+
+## Choose this workflow
+
+Use for evidence-backed removal of unnecessary code or assets; refactor preserves behavior through restructuring.
+
+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; directory/change scope and cleanup objective. Requires references, build configuration, and relevant checks.
+
+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
+
+Demonstrably unused code, stale artifacts, or needless complexity; no behavior redesign.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Find candidates, check dynamic/configuration references and public exports, remove only supported candidates, and verify affected builds/behavior.
+- Check static callers, public exports, framework conventions and runtime registration; remove a coherent candidate set with a restorable diff.
+
+## Decision branches
+
+- **When use cannot be excluded because loading is dynamic:** Retain the candidate and name the missing runtime or configuration evidence.
+
+## Deliver and verify
+
+- Focused deletions/simplifications with evidence of non-use and checks.
+- Removed items with non-use evidence and checks for affected consumers.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Verified dead code is removed; a dynamically loaded module survives despite lacking static imports.
+
+## Stop and recover
+
+- Uncertain candidates remain with an explanation. Never delete user data or untracked files merely because they look temporary.
+
+## Example requests
+
+- **Normal (apply):** Remove demonstrably unused checkout helpers; preserve public exports.
+- **edge (apply):** Clean dead files while preserving route files discovered by filename.
+- **blocked (inspect):** Identify cleanup candidates without build access; leave uncertain dynamic modules intact.

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

@@ -0,0 +1,56 @@
+---
+name: compare
+description: "Compare specific implementation approaches and their tradeoffs Use for a factual side-by-side comparison; decide recommends adoption and decision-matrix handles weighted priorities."
+---
+
+# compare
+
+Compare specific implementation approaches and their tradeoffs
+
+## Choose this workflow
+
+Use for a factual side-by-side comparison; decide recommends adoption and decision-matrix handles weighted priorities.
+
+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; named approaches, project requirements, and important tradeoffs. Requires enough evidence to evaluate each option.
+
+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
+
+Compare concrete alternatives; weighted decision policy belongs to `decision-matrix`.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Normalize assumptions, compare behavior, complexity, maintenance, migration, and relevant cost; identify where evidence is missing.
+- Normalize workload, feature requirements and time horizon; include a baseline/current option and distinguish switching cost from steady-state cost.
+
+## Decision branches
+
+- **When options meet different hard requirements:** Eliminate infeasible choices before ranking and explain the decisive incompatibility.
+
+## Deliver and verify
+
+- Side-by-side comparison, conditional recommendation, and a discriminating test if needed.
+- Comparable assumptions, evidence-backed differences and conditions that change the result.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Options solve the same stated problem; a hard compatibility constraint can disqualify an otherwise attractive option.
+
+## Stop and recover
+
+- Do not invent benchmark or cost figures. If requirements are unresolved, give conditional choices rather than an arbitrary winner.
+
+## Example requests
+
+- **Normal (inspect):** Compare queue-backed jobs with our existing database job runner.
+- **edge (inspect):** Compare two queues when ordering is required only within an account.
+- **blocked (inspect):** Compare these proposals without usage or pricing data; leave costs bounded or unknown.

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

@@ -0,0 +1,56 @@
+---
+name: copy
+description: "Improve interface wording and product messaging Use for interface language preserving product semantics; docs explains implementation and usage."
+---
+
+# copy
+
+Improve interface wording and product messaging
+
+## Choose this workflow
+
+Use for interface language preserving product semantics; docs explains implementation and usage.
+
+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 to specified UI/docs; plan for alternatives only. Requires audience, intent, voice, and relevant product facts.
+
+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
+
+Labels, help text, error messages, and supplied marketing content; no invented guarantees or policy changes.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Identify user decisions, preserve meaning, make actions/errors specific, check space/localization constraints, and update authorized surfaces.
+- Identify the decision each label or message supports; preserve legal/business meaning and test truncation, pluralization and missing-value variants.
+
+## Decision branches
+
+- **When improving clarity would change a product promise:** Surface that policy choice rather than quietly rewriting it.
+
+## Deliver and verify
+
+- Revised copy with necessary context and affected states.
+- Copy changes by state, rationale, character constraints and unresolved policy wording.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A failure message gives an actionable next step; labels remain understandable without surrounding visual context.
+
+## Stop and recover
+
+- Flag unverifiable product claims. Preserve legal or contractual wording unless the requested scope includes changing it.
+
+## Example requests
+
+- **Normal (apply):** Make payment failure messages actionable and preserve the user's entered data.
+- **edge (apply):** Improve payment errors while preserving a user's entered form values.
+- **blocked (inspect):** Suggest copy with unknown refund policy; do not invent eligibility promises.

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

@@ -0,0 +1,56 @@
+---
+name: coverage
+description: "Identify important untested behaviors and prioritize them Use to prioritize missing behavioral checks; test implements the selected cases."
+---
+
+# coverage
+
+Identify important untested behaviors and prioritize them
+
+## Choose this workflow
+
+Use to prioritize missing behavioral checks; test implements the selected cases.
+
+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; subsystem/change and existing test or coverage 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
+
+Important missing behavioral coverage; no automatic test generation or percentage chasing.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Map requirements and failure paths to tests, inspect assertions rather than names, and rank gaps by consequence and likelihood.
+- Trace important failure and recovery paths to actual assertions; look for tests that pass when the requirement is deliberately violated.
+
+## Decision branches
+
+- **When line coverage is high but assertions ignore the effect:** Prioritize the missing invariant over increasing the percentage.
+
+## Deliver and verify
+
+- Prioritized test opportunities with suggested layer, setup, and expected assertion.
+- Ranked gap table with consequence, setup, layer and expected assertion.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Untested recovery behavior is identified despite high line coverage; a low-risk uncovered getter is not prioritized over access control.
+
+## Stop and recover
+
+- Distinguish measured coverage from inferred coverage. Running new coverage jobs requires execution authorization.
+
+## Example requests
+
+- **Normal (inspect):** Identify the highest-risk untested billing behaviors from existing tests.
+- **edge (inspect):** Audit coverage of failed checkout retries despite 95 percent line coverage.
+- **blocked (inspect):** Inspect test sources without a coverage report; avoid claiming measured percentages.

package/plugins/just-vibe/skills/data-backfill/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: data-backfill
+description: "Plan or run resumable historical-data backfills Use for bounded historical reprocessing; data-pipeline creates normal transformation behavior."
+---
+
+# data-backfill
+
+Plan or run resumable historical-data backfills
+
+## Choose this workflow
+
+Use for bounded historical reprocessing; data-pipeline creates normal transformation behavior.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Data engineering methods](../../references/packs/data.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; historical range, transformation/version, target, batch limits, and load budget.
+
+data source/version, schema/semantics, transformation code, permitted sampling scope, and storage/compute budget. Prefer aggregates and redacted samples; never upload datasets to external services implicitly. Record time zones and snapshot identity for reproducibility.
+
+Declared evidence requirements: `data.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Resumable historical recomputation; execution requires the specified data/environment authorization.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Estimate volume, partition work, define idempotent writes and checkpoints, validate a small authorized batch, reconcile output, and resume within limits.
+- Partition a fixed source snapshot, define idempotent writes and checkpoints, measure a small permitted batch and account for concurrent incremental writers.
+
+## Decision branches
+
+- **When a batch fails or load exceeds the cap:** Pause with its partition identity and reconciliation state so resume cannot duplicate or overwrite good output.
+
+## Deliver and verify
+
+- Backfill plan/script or run record with progress, discrepancies, and recovery steps.
+- Partition plan, caps, checkpoints, reconciled counts and resume instructions.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Restart skips or safely replays completed batches; live incremental processing does not race into inconsistent output.
+
+## Stop and recover
+
+- Pause on load/error thresholds. Never launch an unbounded production scan or overwrite historical data without explicit scope.
+
+## Example requests
+
+- **Normal (plan):** Plan a resumable one-year backfill with explicit batch and load limits.
+- **edge (plan):** Resume a backfill while the live pipeline updates the same historical records.
+- **blocked (inspect):** Plan a backfill without production scan permission or known source volume.

package/plugins/just-vibe/skills/data-contract/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: data-contract
+description: "Define schema, semantics, freshness, and quality constraints Use to define producer/consumer data expectations; data-quality checks an accepted contract."
+---
+
+# data-contract
+
+Define schema, semantics, freshness, and quality constraints
+
+## Choose this workflow
+
+Use to define producer/consumer data expectations; data-quality checks an accepted contract.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Data engineering methods](../../references/packs/data.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; producer/consumer needs, schema, field meaning, freshness, and quality requirements.
+
+data source/version, schema/semantics, transformation code, permitted sampling scope, and storage/compute budget. Prefer aggregates and redacted samples; never upload datasets to external services implicitly. Record time zones and snapshot identity for reproducibility.
+
+Resolve any task-specific tools, target identity and evidence before dependent actions. No external connection is assumed.
+
+## Scope
+
+Machine-checkable and semantic obligations across a data boundary.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Identify required fields and keys, define ranges/nullability/time semantics, set freshness expectations, and specify versioning and violation handling.
+- Specify grain, keys, types, units, nullability, event/arrival time, freshness and allowed schema evolution from actual consumption paths.
+
+## Decision branches
+
+- **When thresholds or ownership have not been agreed:** Mark them proposed with rationale and name the decision needed before enforcement.
+
+## Deliver and verify
+
+- Data contract, examples, validation rules, and ownership questions.
+- Versioned field/rule/response contract with accepted and rejected examples.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Syntactically valid but semantically invalid values can fail; an additive schema change has defined compatibility behavior.
+
+## Stop and recover
+
+- Do not invent quality thresholds or ownership agreement. Unresolved thresholds remain proposed values with rationale.
+
+## Example requests
+
+- **Normal (plan):** Define schema, event-time semantics, freshness, and violation handling for orders.
+- **edge (plan):** Define a contract for late-arriving corrections and optional new fields.
+- **blocked (inspect):** Draft a contract without agreed freshness thresholds; do not invent producer commitments.

package/plugins/just-vibe/skills/data-incremental/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: data-incremental
+description: "Implement checkpoints, deduplication, and incremental processing Use for replay-safe CDC or watermark processing; data-backfill handles bounded historical ranges."
+---
+
+# data-incremental
+
+Implement checkpoints, deduplication, and incremental processing
+
+## Choose this workflow
+
+Use for replay-safe CDC or watermark processing; data-backfill handles bounded historical ranges.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Data engineering methods](../../references/packs/data.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; source change mechanism, stable keys, watermark/checkpoint, late-data policy, and destination semantics.
+
+data source/version, schema/semantics, transformation code, permitted sampling scope, and storage/compute budget. Prefer aggregates and redacted samples; never upload datasets to external services implicitly. Record time zones and snapshot identity for reproducibility.
+
+Declared evidence requirements: `data.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Incremental updates, deduplication, deletes, and resume behavior.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Define ordering and checkpoint transactions, handle overlap/late arrivals, make replay safe, and test crashes around write/checkpoint boundaries.
+- Define event versus arrival order, stable keys, deletions and overlap; persist checkpoints only after durable effects and test both sides of that boundary.
+
+## Decision branches
+
+- **When late updates fall behind the current watermark:** Use an explicit overlap/reconciliation policy rather than silently skipping them.
+
+## Deliver and verify
+
+- Incremental processor with state format and restart evidence.
+- Watermark/key protocol and replay, late-update, deletion and crash tests.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Replaying a completed window does not duplicate effects; late updates and deletions follow documented rules.
+
+## Stop and recover
+
+- Do not advance checkpoints before durable output. Missing stable identity or change semantics requires a design decision before implementation.
+
+## Example requests
+
+- **Normal (apply):** Implement checkpointed updates that handle late records and deletion events.
+- **edge (apply):** Process late corrections while safely replaying the previous hour.
+- **blocked (inspect):** Review incremental design without stable source identity; identify the blocking semantic gap.

package/plugins/just-vibe/skills/data-lineage/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: data-lineage
+description: "Trace field origins and transformations Use to trace a field's origin and transformation; trace follows an execution instance."
+---
+
+# data-lineage
+
+Trace field origins and transformations
+
+## Choose this workflow
+
+Use to trace a field's origin and transformation; trace follows an execution instance.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Data engineering methods](../../references/packs/data.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; field/table/report and source/transformation definitions.
+
+data source/version, schema/semantics, transformation code, permitted sampling scope, and storage/compute budget. Prefer aggregates and redacted samples; never upload datasets to external services implicitly. Record time zones and snapshot identity for reproducibility.
+
+Declared evidence requirements: `data.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Origins, transformations, joins, filters, and downstream dependencies for the target data.
+
+None by default. Plan artifacts may be saved when requested.
+
+## Execute
+
+- Trace field expressions through jobs/views, identify version and ownership boundaries, record lossy transformations, and mark opaque external steps.
+- Follow expressions through joins, filters, aggregations and versioned jobs; record grain changes and lossy transformations at each boundary.
+
+## Decision branches
+
+- **When an imported dataset has opaque provenance:** Stop confirmed lineage at that source and request its producer contract rather than assigning an invented origin.
+
+## Deliver and verify
+
+- Field-level lineage graph/table with evidence links and gaps.
+- Field-level path with transformations, versions, owners and opaque boundaries.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- Derived aggregates include filter/join semantics; an opaque imported table is shown as an unknown boundary rather than assigned invented provenance.
+
+## Stop and recover
+
+- Cap expansion at the requested scope. Documentation-only lineage is labeled separately from code-verified lineage.
+
+## Example requests
+
+- **Normal (inspect):** Trace invoice_total through the transforms and source columns.
+- **edge (inspect):** Trace a revenue metric through currency conversion and filtered joins.
+- **blocked (inspect):** Map lineage from partial job definitions without upstream access.

package/plugins/just-vibe/skills/data-pipeline/SKILL.md

@@ -0,0 +1,56 @@
+---
+name: data-pipeline
+description: "Build ingestion or transformation with observable failures Use for a transformation pipeline; data-incremental focuses on checkpoints and change processing."
+---
+
+# data-pipeline
+
+Build ingestion or transformation with observable failures
+
+## Choose this workflow
+
+Use for a transformation pipeline; data-incremental focuses on checkpoints and change processing.
+
+Read [shared execution](../../references/execution.md) for context/mode/authority handling and [Data engineering methods](../../references/packs/data.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; sources, transforms, destination, cadence, correctness criteria, and resource limits.
+
+data source/version, schema/semantics, transformation code, permitted sampling scope, and storage/compute budget. Prefer aggregates and redacted samples; never upload datasets to external services implicitly. Record time zones and snapshot identity for reproducibility.
+
+Declared evidence requirements: `data.read`. Use actual host discovery or adequate supplied artifacts; unavailable evidence remains blocked/unknown.
+
+## Scope
+
+Ingestion/transformation implementation and isolated validation; activating production schedules requires that request.
+
+Only the requested local changes; external actions require their exact action and target in session authorization.
+
+## Execute
+
+- Define source identity and keys, validate inputs, implement transformations and atomic/staged writes, expose failures, and test restart and bad-record handling.
+- Establish stable source/output identity, validate transformations with small hand-checked fixtures and stage writes so completion markers follow durable output.
+
+## Decision branches
+
+- **When invalid records can be isolated without corrupting the batch:** Quarantine with counts/reasons under the declared policy; never drop them silently.
+
+## Deliver and verify
+
+- Pipeline, configuration, quality checks, and operational instructions.
+- Transform mapping, input/output reconciliation, completion protocol and failure accounting.
+
+Verify these observable conditions when applicable to the actual task; do not claim they were exercised from merely reading this file:
+
+- A normal batch produces reconciled output; interrupted writes do not masquerade as a completed partition.
+
+## Stop and recover
+
+- No silent row dropping or unrequested data export. Missing semantics block affected transformations rather than guessed conversions.
+
+## Example requests
+
+- **Normal (apply):** Implement isolated ingestion with observable failures and atomic partition writes.
+- **edge (apply):** Build a pipeline interrupted between writing data and publishing its manifest.
+- **blocked (inspect):** Design transformations with missing field semantics; block only the affected conversions.