video-library-pipeline

This skill is the orchestrator for video-library-pipeline.

Use one of the stage skills below for actual execution:

• •extensions/video-library-pipeline/skills/normalize-review/SKILL.md
• •extensions/video-library-pipeline/skills/extract-review/SKILL.md
• •extensions/video-library-pipeline/skills/move-review/SKILL.md

Rule

• •Use plugin tools, not direct script calls.
• •Execute in the main agent turn; do not delegate pipeline execution to subagents.
• •Never substitute plugin tools with exec shell commands.
  • •video_pipeline_* names are tool names, not shell commands.
  • •If a tool call cannot be issued in the current environment, stop and report it as "tool registry / permissions issue".
• •Health and diagnostics tools:
  • •video_pipeline_validate
  • •video_pipeline_logs
  • •video_pipeline_status
• •Before running, always check path config mismatch risk (sourceRoot, destRoot, windowsOpsRoot) and ask the user to confirm if there is any possibility of wrong path settings.
• •This plugin does not use pnpm test / scripts/*.sh style E2E. E2E is performed by tool-call sequence.

Mandatory interactive flow (human review required)

For user-driven runs, follow this order and stop at each review gate:

1. •Normalization + human review
2. •Extraction + human review
3. •Move/apply + human review

Critical requirement:

• •After extraction, save program info to YAML and review it with the user before move/apply.
• •Use video_pipeline_export_program_yaml to generate this YAML.

Command naming guardrail (required)

• •Do not invent CLI commands such as openclaw video_pipeline_run or openclaw tool ....
• •In this plugin, execution is done by tool calls:
  • •video_pipeline_validate
  • •video_pipeline_analyze_and_move_videos
  • •video_pipeline_logs
  • •video_pipeline_status
  • •video_pipeline_reextract
  • •video_pipeline_apply_reviewed_metadata
  • •video_pipeline_export_program_yaml
  • •video_pipeline_repair_db
• •The only plugin CLI helper command is:
  • •openclaw video-pipeline-status

Exec fallback guardrail (required)

• •Do not run DB checks with python -c via exec for this plugin flow.
• •Do not retry the same failing exec command repeatedly.
• •If command construction fails once due to quoting/syntax (unexpected EOF, unrecognized token, SyntaxError), stop immediately and report:
  • •failing command (brief)
  • •root cause (quote/syntax break)
  • •required fix (use plugin tool call directly)

Path sanity prompt (required)

Before video_pipeline_validate or pipeline execution, ask a short confirmation when path mismatch is possible:

• •"パス設定ミスの可能性があります。sourceRoot / destRoot / windowsOpsRoot は実在パスですか？"
• •If an error says ... does not exist, first treat it as config/path mismatch and ask for the intended real path.

Non-interactive automation flow (cron)

1. •Validate environment:
   • •Call video_pipeline_validate with {"checkWindowsInterop": true}
   • •Stop if ok is false.
2. •Run pipeline:
   • •Call video_pipeline_analyze_and_move_videos with:
     • •apply: true for real move, false for dry-run
     • •maxFilesPerRun: e.g. 500 in cron
     • •allowNeedsReview: default false
3. •Parse run summary:
   • •Parse JSON from tool result stdout.
   • •Expected keys:
     • •inventory
     • •queue
     • •plan
     • •applied
     • •remaining_files
     • •plan_stats
4. •Collect latest pointers:
   • •Call video_pipeline_logs with {"kind":"all","tail":50}.

E2E execution contract (required)

• •For E2E requests, execute this exact order without asking for extra command details:
  1. •video_pipeline_validate with {"checkWindowsInterop": true}
  2. •video_pipeline_analyze_and_move_videos with {"apply": false, "maxFilesPerRun": 1, "allowNeedsReview": false} for first dry-run check
  3. •video_pipeline_logs with {"kind":"all","tail":50}
• •If step 1 cannot be called as a tool (tool not visible/available), do not replace it with shell probing. Stop and report the environment mismatch.
• •Do not use sessions_spawn/subagent for this flow. Keep tool calls in the same main-agent turn.
• •Success can be declared only if all 3 tool calls above produced concrete toolResult outputs in the same session.
• •A completion notice with Findings: (no output) is not a successful E2E result.
• •If step 1 fails, report the failing checks and stop.
• •If step 2 fails, report exitCode, stderr summary, and log pointers.

Alert guidance (cron/automation)

Alert if any of:

• •validate ok is false
• •runner/tool exit code is non-zero
• •summary JSON parse fails
• •remaining_files > 0
• •plan_stats.skipped_needs_review > 0
• •plan_stats.skipped_missing_fields > 0
• •plan_stats.skipped_outside > 0
• •summary/log pointers missing (inventory, queue, plan, applied)

On healthy run, avoid noisy reports.