batchwizard 0.4.0__tar.gz → 0.5.0__tar.gz

batchwizard-0.5.0/.github/workflows/ci.yml

@@ -0,0 +1,52 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+permissions:
+  contents: read
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7.0.0
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+      - uses: astral-sh/setup-uv@v8.3.2
+      - run: uv sync --frozen --group dev
+      - run: uv run --frozen ruff check .
+      - run: uv run --frozen ruff format --check .
+
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.11", "3.12", "3.13", "3.14"]
+    steps:
+      - uses: actions/checkout@v7.0.0
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+      - uses: astral-sh/setup-uv@v8.3.2
+        with:
+          python-version: ${{ matrix.python-version }}
+      - run: uv sync --frozen --group dev
+      - run: uv run --frozen pytest tests/
+
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7.0.0
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+      - uses: astral-sh/setup-uv@v8.3.2
+      - run: uv build
+      - uses: actions/upload-artifact@v7.0.1
+        with:
+          name: dist
+          path: dist/

{batchwizard-0.4.0 → batchwizard-0.5.0}/.github/workflows/publish.yml

@@ -4,16 +4,20 @@ on:
   release:
     types: [published]
 
+permissions:
+  contents: read
+
 jobs:
   build:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7.0.0
         with:
           fetch-depth: 0
-      - uses: astral-sh/setup-uv@v5
+          persist-credentials: false
+      - uses: astral-sh/setup-uv@v8.3.2
       - run: uv build
-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@v7.0.1
         with:
           name: dist
           path: dist/
@@ -26,7 +30,7 @@ jobs:
       # Required for PyPI Trusted Publishing (OIDC) — no API token needed
       id-token: write
     steps:
-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@v8.0.1
         with:
           name: dist
           path: dist/

batchwizard-0.5.0/PKG-INFO

@@ -0,0 +1,205 @@
+Metadata-Version: 2.4
+Name: batchwizard
+Version: 0.5.0
+Summary: Durable batch orchestration across OpenAI and Anthropic
+Project-URL: Homepage, https://github.com/cmakafui/batchwizard
+Project-URL: Repository, https://github.com/cmakafui/batchwizard
+Project-URL: Issues, https://github.com/cmakafui/batchwizard/issues
+Author: Carl Kugblenu
+License-Expression: MIT
+License-File: LICENSE
+Keywords: anthropic,async,batch,cli,openai
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Requires-Python: >=3.11
+Requires-Dist: anthropic[aiohttp]<1,>=0.117
+Requires-Dist: loguru>=0.7
+Requires-Dist: openai[aiohttp]<3,>=2.46
+Requires-Dist: pydantic-settings>=2.3
+Requires-Dist: pydantic>=2.8
+Requires-Dist: python-dotenv>=1.0
+Requires-Dist: rich>=13.7
+Requires-Dist: typer>=0.12
+Description-Content-Type: text/markdown
+
+# BatchWizard
+
+BatchWizard is one durable CLI for OpenAI and Anthropic batch jobs. Submit
+provider-native JSONL, close the terminal, and come back later: a local SQLite
+manifest remembers what is running, what still needs to be downloaded, and what
+needs your attention.
+
+It deliberately does not translate prompts between APIs. OpenAI and Anthropic
+have different request formats and capabilities; BatchWizard gives them one
+operational lifecycle without pretending they are the same protocol.
+
+## Quickstart
+
+Install BatchWizard as an isolated Python 3.11+ tool:
+
+```bash
+uv tool install batchwizard
+```
+
+Provide one or both standard API-key environment variables:
+
+```bash
+export OPENAI_API_KEY="..."
+export ANTHROPIC_API_KEY="..."
+```
+
+Submit native JSONL to either provider:
+
+```bash
+batchwizard submit openai.jsonl --endpoint /v1/responses
+batchwizard submit --provider anthropic anthropic.jsonl
+```
+
+The submit process can exit once the providers accept the work. One later
+command resumes every actionable job, grouped by provider:
+
+```bash
+batchwizard watch --output-directory ./results
+batchwizard status --all
+```
+
+`watch` is intentionally provider-free. The manifest records which adapter owns
+each batch, so a single invocation can resume a mixture of OpenAI and Anthropic
+jobs. Missing credentials or a temporary download failure for one provider do
+not erase work or prevent other provider groups from advancing.
+
+## Input formats
+
+BatchWizard accepts each provider's native JSONL rather than maintaining a lossy
+common prompt schema.
+
+| | OpenAI | Anthropic |
+| --- | --- | --- |
+| Select with | default, or `--provider openai` | `--provider anthropic` |
+| JSONL row | `{"custom_id", "method", "url", "body"}` | `{"custom_id", "params"}` |
+| Model location | `body.model` | `params.model` |
+| Submission | file upload, then Batch creation | requests sent inline to Message Batches |
+| Provider status | Batch `status` | Message Batch `processing_status` |
+| Results | provider output and error files | unordered result stream split by outcome |
+| Recovery after an uncertain submit | automatic intent matching through Batch metadata | manual attachment by batch ID |
+
+### OpenAI
+
+```jsonl
+{"custom_id":"ticket-1","method":"POST","url":"/v1/responses","body":{"model":"gpt-5.4","input":"Classify this ticket as billing, technical, or other."}}
+```
+
+The endpoint passed to BatchWizard must match the `url` in every row:
+
+```bash
+batchwizard submit openai.jsonl --endpoint /v1/responses
+```
+
+`/v1/chat/completions` is the default. BatchWizard also accepts the Responses,
+Embeddings, Completions, Moderations, Images, and Videos Batch API endpoints.
+
+### Anthropic
+
+```jsonl
+{"custom_id":"ticket-1","params":{"model":"claude-opus-4-8","max_tokens":256,"messages":[{"role":"user","content":"Classify this ticket as billing, technical, or other."}]}}
+```
+
+```bash
+batchwizard submit --provider anthropic anthropic.jsonl
+```
+
+BatchWizard validates the batch envelope, `custom_id` syntax and uniqueness,
+batch size, `max_tokens`, and the non-streaming requirement before submission.
+Anthropic remains responsible for validating its evolving Messages parameter
+surface. See [Anthropic Message Batches](docs/anthropic.md) for result routing,
+prompt caching, cancellation, and retention behavior.
+
+## Durable jobs and recovery
+
+The manifest separates remote execution from local artifact collection. A
+provider can report a batch complete while its result files are still pending
+locally; that row remains actionable until collection succeeds or the provider
+confirms that the artifacts are no longer available.
+
+Submission is durable too. BatchWizard writes an intent before making provider
+requests. If a connection fails after the remote service may have accepted the
+batch, it keeps that intent instead of blindly submitting a duplicate:
+
+```bash
+batchwizard reconcile
+batchwizard reconcile INTENT
+batchwizard reconcile INTENT --batch-id MSGBATCH_ID
+```
+
+OpenAI intents can be matched through Batch metadata. Anthropic does not expose
+equivalent batch metadata, so its uncertain submissions are attached after the
+matching batch is identified with `list-jobs`.
+
+The lifecycle and SQLite migrations are documented in
+[Job lifecycle](docs/job-lifecycle.md).
+
+## Result files
+
+Provider output stays JSONL and is written atomically:
+
+- `<batch_id>_results.jsonl` contains successful rows.
+- `<batch_id>_errors.jsonl` contains request-level failures when present.
+
+Anthropic result rows can arrive in any order. BatchWizard preserves each raw row
+and routes `succeeded` results to the results file and `errored`, `expired`, or
+`canceled` results to the errors file.
+
+## Credentials and configuration
+
+Environment variables are the preferred way to supply credentials. Keys can
+also be persisted for convenience:
+
+```bash
+batchwizard configure --provider openai --set-key "$OPENAI_API_KEY"
+batchwizard configure --provider anthropic --set-key "$ANTHROPIC_API_KEY"
+batchwizard configure --show --provider openai
+```
+
+Persisted keys are plaintext JSON. BatchWizard writes the configuration
+atomically with owner-only file permissions, but it does not use the operating
+system keychain in v0.5. Treat that file as a secret; keyring-backed storage is
+deferred to v0.6.
+
+## Commands
+
+| Command | Purpose |
+| --- | --- |
+| `submit` | Submit files and return after provider acceptance. |
+| `watch` | Resume actionable jobs and collect terminal artifacts. |
+| `process` | Submit, watch, and collect in one invocation. |
+| `status` | Inspect the local SQLite manifest. |
+| `reconcile` | Recover a submission with an uncertain provider outcome. |
+| `list-jobs` | List recent jobs directly from one provider. |
+| `cancel` | Request cancellation for a tracked or provider-native batch ID. |
+| `download` | Collect result artifacts for one batch. |
+| `configure` | Store, inspect, or reset local settings. |
+
+Use `batchwizard <command> --help` for the exact arguments and options.
+
+## Development
+
+```bash
+uv sync --all-groups
+uv run pytest
+uv run ruff format --check .
+uv run ruff check .
+uv build
+```
+
+The supported runtime matrix is Python 3.11 through 3.14. The lockfile is
+committed; update it intentionally with `uv lock --upgrade-package <package>`.
+
+## License
+
+[MIT](LICENSE)

batchwizard-0.5.0/README.md

@@ -0,0 +1,175 @@
+# BatchWizard
+
+BatchWizard is one durable CLI for OpenAI and Anthropic batch jobs. Submit
+provider-native JSONL, close the terminal, and come back later: a local SQLite
+manifest remembers what is running, what still needs to be downloaded, and what
+needs your attention.
+
+It deliberately does not translate prompts between APIs. OpenAI and Anthropic
+have different request formats and capabilities; BatchWizard gives them one
+operational lifecycle without pretending they are the same protocol.
+
+## Quickstart
+
+Install BatchWizard as an isolated Python 3.11+ tool:
+
+```bash
+uv tool install batchwizard
+```
+
+Provide one or both standard API-key environment variables:
+
+```bash
+export OPENAI_API_KEY="..."
+export ANTHROPIC_API_KEY="..."
+```
+
+Submit native JSONL to either provider:
+
+```bash
+batchwizard submit openai.jsonl --endpoint /v1/responses
+batchwizard submit --provider anthropic anthropic.jsonl
+```
+
+The submit process can exit once the providers accept the work. One later
+command resumes every actionable job, grouped by provider:
+
+```bash
+batchwizard watch --output-directory ./results
+batchwizard status --all
+```
+
+`watch` is intentionally provider-free. The manifest records which adapter owns
+each batch, so a single invocation can resume a mixture of OpenAI and Anthropic
+jobs. Missing credentials or a temporary download failure for one provider do
+not erase work or prevent other provider groups from advancing.
+
+## Input formats
+
+BatchWizard accepts each provider's native JSONL rather than maintaining a lossy
+common prompt schema.
+
+| | OpenAI | Anthropic |
+| --- | --- | --- |
+| Select with | default, or `--provider openai` | `--provider anthropic` |
+| JSONL row | `{"custom_id", "method", "url", "body"}` | `{"custom_id", "params"}` |
+| Model location | `body.model` | `params.model` |
+| Submission | file upload, then Batch creation | requests sent inline to Message Batches |
+| Provider status | Batch `status` | Message Batch `processing_status` |
+| Results | provider output and error files | unordered result stream split by outcome |
+| Recovery after an uncertain submit | automatic intent matching through Batch metadata | manual attachment by batch ID |
+
+### OpenAI
+
+```jsonl
+{"custom_id":"ticket-1","method":"POST","url":"/v1/responses","body":{"model":"gpt-5.4","input":"Classify this ticket as billing, technical, or other."}}
+```
+
+The endpoint passed to BatchWizard must match the `url` in every row:
+
+```bash
+batchwizard submit openai.jsonl --endpoint /v1/responses
+```
+
+`/v1/chat/completions` is the default. BatchWizard also accepts the Responses,
+Embeddings, Completions, Moderations, Images, and Videos Batch API endpoints.
+
+### Anthropic
+
+```jsonl
+{"custom_id":"ticket-1","params":{"model":"claude-opus-4-8","max_tokens":256,"messages":[{"role":"user","content":"Classify this ticket as billing, technical, or other."}]}}
+```
+
+```bash
+batchwizard submit --provider anthropic anthropic.jsonl
+```
+
+BatchWizard validates the batch envelope, `custom_id` syntax and uniqueness,
+batch size, `max_tokens`, and the non-streaming requirement before submission.
+Anthropic remains responsible for validating its evolving Messages parameter
+surface. See [Anthropic Message Batches](docs/anthropic.md) for result routing,
+prompt caching, cancellation, and retention behavior.
+
+## Durable jobs and recovery
+
+The manifest separates remote execution from local artifact collection. A
+provider can report a batch complete while its result files are still pending
+locally; that row remains actionable until collection succeeds or the provider
+confirms that the artifacts are no longer available.
+
+Submission is durable too. BatchWizard writes an intent before making provider
+requests. If a connection fails after the remote service may have accepted the
+batch, it keeps that intent instead of blindly submitting a duplicate:
+
+```bash
+batchwizard reconcile
+batchwizard reconcile INTENT
+batchwizard reconcile INTENT --batch-id MSGBATCH_ID
+```
+
+OpenAI intents can be matched through Batch metadata. Anthropic does not expose
+equivalent batch metadata, so its uncertain submissions are attached after the
+matching batch is identified with `list-jobs`.
+
+The lifecycle and SQLite migrations are documented in
+[Job lifecycle](docs/job-lifecycle.md).
+
+## Result files
+
+Provider output stays JSONL and is written atomically:
+
+- `<batch_id>_results.jsonl` contains successful rows.
+- `<batch_id>_errors.jsonl` contains request-level failures when present.
+
+Anthropic result rows can arrive in any order. BatchWizard preserves each raw row
+and routes `succeeded` results to the results file and `errored`, `expired`, or
+`canceled` results to the errors file.
+
+## Credentials and configuration
+
+Environment variables are the preferred way to supply credentials. Keys can
+also be persisted for convenience:
+
+```bash
+batchwizard configure --provider openai --set-key "$OPENAI_API_KEY"
+batchwizard configure --provider anthropic --set-key "$ANTHROPIC_API_KEY"
+batchwizard configure --show --provider openai
+```
+
+Persisted keys are plaintext JSON. BatchWizard writes the configuration
+atomically with owner-only file permissions, but it does not use the operating
+system keychain in v0.5. Treat that file as a secret; keyring-backed storage is
+deferred to v0.6.
+
+## Commands
+
+| Command | Purpose |
+| --- | --- |
+| `submit` | Submit files and return after provider acceptance. |
+| `watch` | Resume actionable jobs and collect terminal artifacts. |
+| `process` | Submit, watch, and collect in one invocation. |
+| `status` | Inspect the local SQLite manifest. |
+| `reconcile` | Recover a submission with an uncertain provider outcome. |
+| `list-jobs` | List recent jobs directly from one provider. |
+| `cancel` | Request cancellation for a tracked or provider-native batch ID. |
+| `download` | Collect result artifacts for one batch. |
+| `configure` | Store, inspect, or reset local settings. |
+
+Use `batchwizard <command> --help` for the exact arguments and options.
+
+## Development
+
+```bash
+uv sync --all-groups
+uv run pytest
+uv run ruff format --check .
+uv run ruff check .
+uv build
+```
+
+The supported runtime matrix is Python 3.11 through 3.14. The lockfile is
+committed; update it intentionally with `uv lock --upgrade-package <package>`.
+
+## License
+
+[MIT](LICENSE)

batchwizard-0.5.0/docs/anthropic.md

@@ -0,0 +1,78 @@
+# Anthropic Message Batches
+
+BatchWizard uses the GA async Message Batches surface in the official Anthropic
+Python SDK. Inputs and saved result rows stay provider-native; only lifecycle,
+request counts, and local artifact locations are normalized.
+
+## Input
+
+Each nonempty UTF-8 JSONL line must contain a unique `custom_id` and a Messages
+API `params` object:
+
+```jsonl
+{"custom_id":"ticket-42","params":{"model":"claude-opus-4-8","max_tokens":256,"messages":[{"role":"user","content":"Classify this ticket."}]}}
+```
+
+```bash
+batchwizard submit --provider anthropic input.jsonl
+```
+
+BatchWizard validates the batch envelope before making an HTTP request:
+
+- 1–100,000 requests and a serialized request size no greater than 256 MB
+- unique 1–64 character `custom_id` values containing only letters, numbers,
+  hyphens, and underscores
+- object-valued `params` with `model`, `messages`, and `max_tokens >= 1`
+- `stream` must not be `true`
+
+The complete Messages parameter surface evolves independently. Anthropic
+validates those parameters asynchronously and returns invalid rows as
+`result.type == "errored"`; BatchWizard does not claim to reproduce that server
+validator.
+
+### Prompt caching
+
+Message Batches and prompt caching can be combined, and their pricing discounts
+stack. Put the shared prefix first and use identical `cache_control` breakpoints
+across rows. Because rows run asynchronously and concurrently, cache hits are
+best-effort; Anthropic recommends the 1-hour TTL for batches with shared context
+that may process beyond the default 5-minute cache lifetime.
+
+## Lifecycle and outcomes
+
+Anthropic exposes three batch lifecycle statuses:
+
+| Anthropic | BatchWizard |
+| --- | --- |
+| `in_progress` | `running` |
+| `canceling` | `cancelling` |
+| `ended` | terminal `completed` |
+
+`ended` is neutral: the batch may contain any mixture of `succeeded`, `errored`,
+`canceled`, and `expired` requests. Those counters are recorded independently, so
+an all-errored batch is not mislabeled as a provider-level failure.
+
+Cancellation first returns `canceling`. The eventual batch status is `ended`, and
+partial successful results may still exist.
+
+## Results and retention
+
+The SDK streams an unordered JSONL result sequence. BatchWizard preserves every
+native object and routes by row result type:
+
+- `succeeded` → `<batch_id>_results.jsonl`
+- `errored`, `expired`, or `canceled` → `<batch_id>_errors.jsonl`
+
+Both destination files are built as temporary files and exposed only after the
+entire decoder finishes. A network or disk interruption leaves previous final
+files untouched and keeps collection actionable. The decoder is explicitly
+closed on both success and failure.
+
+Anthropic retains results for 29 days from batch creation. Once the provider
+reports them archived, collection becomes `unavailable` rather than retrying a
+permanent loss forever.
+
+See the current [Message Batches guide](https://platform.claude.com/docs/en/build-with-claude/batch-processing),
+[prompt caching guide](https://platform.claude.com/docs/en/build-with-claude/prompt-caching),
+[Python SDK documentation](https://platform.claude.com/docs/en/cli-sdks-libraries/sdks/python),
+and [Batch API reference](https://platform.claude.com/docs/en/api/python/messages/batches).

batchwizard-0.5.0/docs/job-lifecycle.md

@@ -0,0 +1,121 @@
+# Job lifecycle
+
+BatchWizard treats a provider batch and the CLI process watching it as separate
+things. A submitted job remains in the SQLite manifest until its remote lifecycle
+and local artifacts can be inspected independently.
+
+## State dimensions
+
+`JobRecord` deliberately records four different concerns:
+
+| Concern | Manifest fields | Meaning |
+| --- | --- | --- |
+| Remote lifecycle | `state`, `provider_status` | The normalized lifecycle and the provider's exact status string |
+| Request outcomes | `completed_count`, `failed_count`, `cancelled_count`, `expired_count`, `total_count` | Per-request results; a terminal job can contain a mixture |
+| Artifact collection | `collection_state`, `output_path`, `error_path` | Whether all currently available provider artifacts are durable locally |
+| Local health | `last_local_error`, `poll_failures` | Connectivity, authentication, or filesystem problems that must not rewrite remote truth |
+
+The normalized remote states are:
+
+```text
+submitting -> pending -> running -> completed | failed | expired
+                              -> cancelling -> cancelled
+```
+
+Providers may skip states. An unknown provider status is preserved in
+`provider_status`; BatchWizard does not guess whether it is terminal.
+`submitting` is a durable local intent written before provider I/O. It has no
+provider batch ID until submission is confirmed or reconciled.
+
+Artifact collection progresses independently:
+
+```text
+not_ready -> pending -> collected
+                   \-> failed -> pending (on the next retry)
+                   \-> unavailable
+```
+
+`collected` means the collection attempt completed successfully. A provider may
+legitimately have no output or error file for a terminal job.
+`unavailable` means the provider has permanently removed or archived artifacts;
+it is terminal and is not retried.
+
+## Actionable jobs
+
+A job is actionable when either:
+
+1. It is an unresolved `submitting` intent;
+2. Its remote state is `pending`, `running`, or `cancelling`; or
+3. Its remote state is terminal and artifact collection is `pending` or `failed`.
+
+`batchwizard watch` operates on actionable jobs. This is why a temporary download
+failure survives process exit and is retried by a later invocation.
+
+## Failure invariants
+
+- A polling, authentication, rate-limit, SDK, or network error is a local
+  observation failure. It never changes the remote job to `failed`.
+- Retryable polling errors are attempted a bounded number of times in one watch
+  invocation. The job remains actionable after the watcher pauses.
+- A provider-reported terminal status is stored before artifact collection begins.
+- A submission intent is committed before provider I/O. An uncertain submission
+  outcome is never retried as a new batch automatically.
+- Downloads are streamed to a temporary file in the destination directory and
+  atomically renamed only after the stream completes.
+- A failed collection records `collection_state=failed` and remains actionable.
+- A provider-confirmed retention loss records `collection_state=unavailable` and
+  stops retrying without pretending the artifacts were collected.
+- Request-level errors do not change a successfully ended provider job into a
+  provider failure; their counts and error artifact are recorded separately.
+- A cancellation request is not equivalent to cancellation completion. The
+  provider's immediate `cancelling` state remains active until confirmed terminal.
+
+## Provider contract
+
+A provider adapter owns submission mechanics and native statuses, but exposes a
+small operational contract:
+
+- Submit a provider-native input file.
+- Return a normalized status snapshot while preserving the raw status.
+- Fetch all available success and error artifacts idempotently.
+- Request cancellation and return its immediate status.
+- Return provider-neutral summaries for `list-jobs`.
+
+Input and result payloads remain provider-native. BatchWizard normalizes lifecycle
+and artifact locations, not model-specific request semantics.
+
+OpenAI exposes job-level terminal statuses. Anthropic exposes the neutral terminal
+status `ended` and independent `succeeded`, `errored`, `canceled`, and `expired`
+request counts. BatchWizard maps `ended` to a terminal normalized lifecycle while
+preserving those row outcomes; an all-errored Anthropic batch is not rewritten as
+a remote provider failure.
+
+## Schema migrations
+
+The SQLite manifest uses `PRAGMA user_version`. Opening a v0.4 manifest migrates it
+through schema versions 1, 2, and 3:
+
+- Active jobs start with artifact state `not_ready`.
+- Terminal jobs with an existing output or error path become `collected`.
+- Terminal jobs without a known local path become `pending` so collection is
+  conservatively retried.
+- Version 2 makes `(provider, batch_id)` the durable identity. Provider-native IDs
+  no longer have to be globally unique across different providers.
+- Version 3 adds a unique submission intent, makes the provider batch ID and
+  endpoint nullable until confirmation, and preserves existing rows with stable
+  legacy intent IDs.
+
+A manifest with a newer schema version is rejected rather than silently modified
+by an older BatchWizard release.
+
+## Submission reconciliation
+
+Submission crosses the local manifest and a remote provider. BatchWizard writes a
+`submitting` intent first, then updates that same row after the provider returns
+its batch ID. OpenAI receives the intent ID as batch metadata, allowing
+`batchwizard reconcile INTENT` to find and attach a batch after a crash or
+uncertain connection failure. Anthropic does not expose equivalent batch
+metadata; identify the matching job by creation time with `list-jobs`, then run
+`batchwizard reconcile INTENT --batch-id MSGBATCH_ID`. After confirming that no
+provider batch exists, remove the local intent with
+`batchwizard reconcile INTENT --discard`.