Labali Spotify Delete Draft Episodes

This skill uses unified runtime by design: deterministic trajectory cache first, then policy executor fallback for reliability. The deterministic first-level cache script is available at scripts/deterministic.ts with CLI wrapper scripts/run_deterministic.ts.

Required Constraints

• Use browser automation only.
• Do not use Spotify APIs.
• Use semantic interactions first (visible text, label, role).
• Login is manual; session reuse is required.
• Deletion success is validated by business state, not by click success.
• Scope deletions to Draft filter only.

NEVER

• Never confirm deletion success by click outcome alone — verify the Draft list no longer contains the deleted episode.
• Never delete episodes outside the Draft filter scope — only target episodes in the Draft list.
• Never return success before the policy executor's business-state verification passes.

Success Criteria

A run is successful only when all conditions hold:

1. The target show is selected by show_id/show URL.
2. At least one Draft episode is deleted in default mode (delete_all_drafts=false).
3. In full-cleanup mode (delete_all_drafts=true), final check confirms no Draft episodes remain.

Runtime Inputs

Use skill.yaml as the source of truth for input schema.

Operational Mode

• Delete mode:
  • Default: delete_all_drafts=false deletes only the first Draft episode.
  • Full cleanup: set delete_all_drafts=true to delete all Draft episodes (bounded by max_delete).
• Default unified mode: run deterministic trajectory cache first, then auto-downgrade to policy executor.
  • Deterministic trajectory cache (scripts/deterministic.ts) is optional acceleration only.
  • Policy executor (scripts/executor.ts) is mandatory reliability baseline and must succeed independently.
  • Set disable_deterministic_cache=true (or CLI --disable_deterministic_cache true) to skip deterministic mode and run policy executor directly.
• If deterministic trajectory cache fails:
  • continue with policy executor in the same run,
  • record deterministic failure context for later optimization.
• If policy executor fails:
  • prioritize policy repair and retry in a loop until business-success criteria pass,
  • do not return success before verification passes.

If policy executor stage decisions are unclear, load references/architecture.md before proceeding.

• After task completion:
  • use deterministic failure records plus policy-success evidence to optimize deterministic mode incrementally,
  • keep deterministic mode optional; never weaken policy baseline for speed-only changes.

Resources

• Architecture and standards: references/architecture.md
• Workflow map and semantic action plan: references/plan.md
• Unified deterministic->policy entry: scripts/auto-executor.ts
• Executor orchestration entry: scripts/executor.ts
• Deterministic first-level cache entry: scripts/deterministic.ts
• Stage detection module: scripts/stage-detector.ts
• Draft deletion module: scripts/deleter.ts
• Verification module: scripts/verifier.ts
• Shared core runtime module: scripts/core.ts
• CLI wrapper: scripts/run.ts
• Deterministic CLI wrapper: scripts/run_deterministic.ts
• Regression checks: tests/test_regression.sh