feat(skills,pipeline): add cinematic series tooling, per-project manifest manager, and production skills

[tienoho]

Summary

This Merge Request enhances FlowKit with a complete Cinematic AI Series Production Suite, designed for creating long-form, multi-episode films (16:9 Landscape & 9:16 Shorts) using Google Flow and Veo 3.

It introduces 5 new cinematic skills, automated tools for compilation and mastering (chapter bumpers, sub-bass drones, hardsubs), an automated per-project Series Manifest manager, and seamless tool sync across Claude Code, Codex CLI, and Antigravity IDE.

---

Key Changes & Features

1. 🎬 Cinematic AI Film Skills (skills/)

• /fk-scriptwriter: Screenplay bible for AI films with 12 scenes/chapter, 3-phase 8s subclip pacing (0-3s, 3-6s, 6-8s), strict narrator speech math (≤ 7.0s, ~18–22 words), and prompt safety hardening.
• /fk-character-bible: Multi-character consistency framework, role-based aliases, entity code-names, and celebrity/public-figure policy bypass techniques.
• /fk-action-choreography: Cinematic combat choreography, camera tracking dynamics, kinetic motion prompts, and collision framing.
• /fk-cinematic-transitions: Transition design, 0.8s dip-to-black fade curves, chapter bumpers, and inter-scene acoustic bridges.
• /fk-sound-design: Multi-track audio mastering, intelligent background ducking (narration boost +2.9 dB, Veo 3 ambient ducking -12 dB), and low-frequency brown noise drones (40–60 Hz).

2. 🛠️ Automation & Mastering Scripts (scripts/)

• scripts/series_manifest.py: Per-project series manifest manager (output/<slug>/series_manifest.json) that caches shared character/location UUIDs across episodes to eliminate redundant reference generations.
• scripts/build_master_film.py: Automated compiler for multi-chapter films, producing cohesive master videos with interstitial chapter bumpers and cinematic audio balance.
• scripts/generate_subtitles.py: Automated .srt / .vtt timeline alignment and FFmpeg hardsub burner (--burn-subtitles) with 4 aesthetic presets (cinematic_gold, clean_white, cyber_amber, emerald_hud).
• scripts/sfx_layering.py: Action-keyword Foley audio layering and spatial sound design.
• scripts/sync_skills.py & scripts/validate_skills.py: Quality assurance linter enforcing 100% English language compliance, YAML frontmatter parity, and cross-IDE synchronization.

3. ⚙️ Core Engine & Extension Enhancements

• setup.py: Added full support for Antigravity IDE (.agents/skills/), offering an interactive target selector (Claude Code, Codex, Antigravity, or All).
• agent/api/projects.py: Enhanced /manifest endpoint to automatically sync and merge all episode records from the database.
• agent/worker/processor.py: Added a safety guard to prevent polling timeouts from erroneously triggering entity re-uploads.
• agent/services/flow_batch.py: Improved TRPC response parsing window search to avoid false media ID mismatches.
• extension/background.js: Prioritized active /project/ tabs for reCAPTCHA token execution and TRPC batch RPC calls.
• .gitignore: Untracked large local runtime dumps (scratch/captured_netlog.jsonl) and ignored temporary media artifacts.

---

Verification & Testing

• Skill Quality Validator: Passed python scripts/validate_skills.py — 41/41 skills verified with zero warnings, 100% English compliant.
• Tool Parity: Ran python scripts/sync_skills.py — Parity achieved across skills/, .agents/skills/, and .claude/commands/.
• Series Manifest Sync: Verified python scripts/series_manifest.py sync --project-id <PID> successfully registers and links shared entities.
• Live Pipeline Run: Tested full end-to-end production of multi-chapter series (Chapters 1–5 master compiled, Chapters 6–7 rendering concurrently via Flow Batch API).

---

Checklist

• Code follows the repository's architectural patterns and guidelines.
• No breaking changes to existing REST endpoints or provider APIs.
• Sensitive files and heavy media artifacts are ignored in .gitignore.
• Documentation and skills are up-to-date.