StaticFlow CLI Publisher

Use this skill to publish Markdown/blog notes into LanceDB and verify results. It is a publish/verify skill, not a table-repair skill.

When To Use

1. Publish one Markdown article (write-article).
2. Batch import images (write-images) or sync a notes directory (sync-notes).
3. Query/update/upsert data in articles, images, and taxonomies.
4. Update bilingual fields from files (db update-article-bilingual).
5. Backfill article vectors (db backfill-article-vectors).
6. Run backend-equivalent local API queries for verification/debug.
7. Do routine post-batch index coverage only after successful writes on healthy tables.
8. Manually complete a music wish (complete-wish).

Execution Policy (Mandatory)

• Context-first: read article and metadata from current context/local files first.
• LLM-native generation: generate missing metadata and summary directly in-session.
• Do not call external model APIs, proxy scripts, or sub-agent model commands.
• Use CLI only for fetch fallback, write/sync, and verification.
• Keep intermediate artifacts under /tmp/, not in the project root directory.
• Never run destructive operations unless explicitly requested.
• Treat duplicate-key, merge-ambiguity, and scalar-index errors as storage-integrity incidents, not routine publish failures.
• On a live production DB with a running backend, do not improvise repair by chaining delete-rows, drop-index, ensure-indexes, optimize --all --prune-now, cleanup-orphans, or rebuild-table-stable.
• If storage maintenance or table repair is the main task, switch to lancedb-optimize or an explicit repair workflow instead of continuing this skill.

Load Extra Context

• Required: references/publish-checklist.md
• If summary generation is needed: ../article-summary-architect/SKILL.md
• Optional docs: README.md, docs/cli-user-guide.zh.md

Preconditions

1. Resolve CLI in this order:
   • ./target/release/sf-cli
   • ./target/debug/sf-cli
   • ../target/release/sf-cli
   • sf-cli from PATH
2. Verify CLI works: <cli> --help
   • Build if needed: cargo build -p sf-cli --release
3. If the checkout is newer than the chosen binary, rebuild before use.
4. Do not prefer legacy ./bin/sf-cli snapshots for storage-format-sensitive writes.
5. Verify DB path exists.
6. Verify DB tables:
   • <cli> db --db-path <db_path> list-tables
   • required: articles, images, taxonomies
   • current content DB may also include runtime tables such as article_views, api_behavior_events, article_requests*, and interactive_*
7. If DB is uninitialized, ask user before:
   • <cli> init --db-path <db_path>

Hard Stop Conditions (Mandatory)

Stop the normal publish workflow immediately if any of these appear:

1. Ambiguous merge inserts
2. RowAddrTreeMap::from_sorted_iter called with non-sorted input
3. Duplicate taxonomy/article key errors that persist after input-side dedupe
4. Full-list reads succeed but filtered single-row reads fail on a long-running backend
5. Any sign that the target table/index layout is already unhealthy

Required response:

1. Stop retries and stop mutating the affected tables.
2. Capture diagnostics first:
   • <cli> db --db-path <db_path> audit-storage --table <table>
   • <cli> db --db-path <db_path> list-indexes <table> --with-stats
   • targeted query-rows / count-rows
3. Preserve rollback state before repair.
4. Escalate to a stable-row-id rebuild / temp-DB rewrite workflow.
5. If a table directory is rebuilt or swapped while backend is still running, assume the old backend may keep a stale manifest/index view; validate with a fresh process and plan blue-green restart before switching traffic.

Publication Workflow

A) Single Article (write-article)

1. Read Markdown body/frontmatter.
2. Ensure required metadata in final payload:
   • summary
   • tags
   • category
   • category_description
   • when bilingual publish is requested: content_en and detailed_summary.zh/en
3. Metadata priority:
   • frontmatter
   • explicit user CLI args
   • skill inference from content + existing taxonomy records
4. Preserve source date:
   • use --date when user provides it
   • else keep frontmatter date
   • else derive from file birth time/mtime (YYYY-MM-DD) and pass --date
5. Summary gate (before publish):
   • regenerate detailed_summary.zh/en when missing/stale/unstructured
   • follow article-summary-architect quality contract
6. Publish command:
   • frontmatter complete:
     • <cli> write-article --db-path <db_path> --file <post.md>
   • frontmatter incomplete:
     • <cli> write-article --db-path <db_path> --file <post.md> --summary "..." --tags "a,b" --category "..." --category-description "..."
   • with custom id:
     • --id <custom_id> (defaults to markdown file stem)
   • with explicit date:
     • --date YYYY-MM-DD
   • explicit bilingual files (preferred for non-frontmatter workflow):
     • <cli> write-article --db-path <db_path> --file <post.md> --summary "..." --tags "a,b" --category "..." --category-description "..." --content-en-file <content_en.md> --summary-zh-file <summary_zh.md> --summary-en-file <summary_en.md>
   • with pre-computed vectors:
     • --vector <json_array> / --vector-en <json_array> / --vector-zh <json_array>
   • with auto-embedding language hint:
     • --language en|zh
   • disable auto-optimize after write:
     • --no-auto-optimize
   • --no-auto-optimize is only for healthy-table publish batching; it is not a repair switch
7. Local image import (optional):
   • --import-local-images
   • optional --media-root <path> (repeatable)
   • optional --generate-thumbnail --thumbnail-size <n>
   • supports ![](path), ![[path]], ![[path|alias]]
8. Verify publication:
   • article row exists
   • taxonomy rows updated
   • images imported when expected
   • sf-cli api get-article <id> returns the updated row on a fresh CLI connection
   • if any verification step throws merge/index integrity errors, stop instead of attempting ad-hoc maintenance
9. Report:
   • article id
   • inferred metadata (if any)
   • summary generated/reused
   • bilingual fields generated/reused (content_en, detailed_summary.zh/en)
   • image import count and warnings

B) Image Batch (write-images)

<cli> write-images --db-path <db_path> --dir <image_dir> [--recursive] [--generate-thumbnail] [--thumbnail-size <n>] [--no-auto-optimize]

Storage note:

• images.data is blob v2 in current production layout
• images.thumbnail remains regular Binary
• prefer verifying image readability through <cli> api get-image ... instead of assuming raw Arrow Binary

C) Notes Sync (sync-notes)

<cli> sync-notes --db-path <db_path> --dir <notes_dir> [--recursive] [--generate-thumbnail] [--thumbnail-size <n>] [--language <en|zh>] [--default-category <name>] [--default-author <name>] [--no-auto-optimize]

DB and API Quick Map

• Safe publish/debug ops: <cli> db --db-path <db_path> <subcommand>
  • table info: list-tables, describe-table <table>, count-rows <table> [--where ...]
  • targeted data checks: query-rows, update-rows, list-indexes <table> [--with-stats]
  • upsert: upsert-article --json <json>, upsert-image --json <json>
  • bilingual: update-article-bilingual --id <id> [--content-en-file ...] [--summary-zh-file ...] [--summary-en-file ...]
  • vectors: backfill-article-vectors [--limit N] [--dry-run]
  • special: reembed-svg-images [--limit N] [--dry-run]
  • schema: create-table <table>
• Escalation-only repair/maintenance ops:
  • delete-rows
  • ensure-indexes, drop-index <name>
  • optimize, cleanup-orphans [--table <table>]
  • rebuild-table-stable <table>, migrate-images-blob-v2, drop-table <table> --yes
  • only use these when the user explicitly asked for DB repair/maintenance and you have a rollback and verification plan
• API-equivalent ops: <cli> api --db-path <db_path> <subcommand>
  • articles: list-articles [--tag ...] [--category ...], get-article <id>, related-articles <id>
  • search: search --q "...", semantic-search --q "..."
  • taxonomy: list-tags, list-categories
  • images: list-images, get-image <id-or-filename>, search-images --id <id>, search-images-text --q "..."
• Legacy query shortcut: <cli> query --table <table> [--where ...] [--columns ...] [--limit N] [--offset N] [--format table|vertical]

Healthy-Table Post-Batch Maintenance

Only after the writes have already succeeded and the affected tables are healthy:

• If you intentionally used --no-auto-optimize for batching, finish with:
  • <cli> db --db-path <db_path> optimize articles
  • <cli> db --db-path <db_path> optimize images
• Use cleanup-orphans / optimize --all --prune-now only as explicit storage maintenance, not as write-failure recovery.
• If compaction/prune is the main task, use the lancedb-optimize skill.

Error Handling

• Missing metadata: infer and continue, but report inferred fields explicitly.
• Image resolution warnings: rerun with --media-root.
• Partial success: rerun only failed steps, then verify again.
• Integrity symptoms (Ambiguous merge inserts, non-sorted input, repeated duplicate keys, stale-backend manifest mismatch) are stop conditions, not retry conditions.
• Never use delete-rows, drop-index, ensure-indexes, optimize, cleanup-orphans, or rebuild-table-stable as an ad-hoc response to a failed write on a live production DB.
• Never run drop-table or delete-rows --all unless user explicitly requests it.