Skip to content

Open-source AI podcast clipper. Generate vertical clips with face tracking and burned-in captions. CLI, MCP server, web app.

feat: add /auto slash command — one-verb pipeline with strategy gate

Nika Siradze7a1557b3f8ef
diff --git a/.claude/commands/auto.md b/.claude/commands/auto.md
new file mode 100644
index 0000000..63d25ad
--- /dev/null
+++ b/.claude/commands/auto.md
@@ -0,0 +1,129 @@
+---
+description: One-verb pipeline — drop a video, confirm strategy, render clips
+allowed-tools: Read, Bash, mcp__podcli__transcribe_podcast, mcp__podcli__get_ui_state, mcp__podcli__set_video, mcp__podcli__suggest_clips, mcp__podcli__batch_create_clips, mcp__podcli__knowledge_base, mcp__podcli__clip_history
+argument-hint: [video-path-or-episode-slug] [optional: count e.g. "5 clips"]
+triggers:
+ - auto
+ - make shorts from this
+ - just edit this
+ - one-shot this
+---
+
+# /auto — One-Verb Pipeline
+
+> Drop raw footage, confirm a strategy, get rendered clips back. Inspired by video-use: "LLM reads, doesn't watch."
+
+This command orchestrates the existing MCP tools on top of the compact packed transcript. No menus. No preset selection. Strategy gate before any render.
+
+---
+
+## Rules
+
+1. **Read, don't watch.** Reason about clips from the packed markdown view — not raw segments, not frame dumps.
+2. **Strategy first, render after.** Propose the cut list and WAIT for user confirmation before calling `batch_create_clips`.
+3. **Knowledge base is context, not template.** If `.podcli/knowledge/` exists, read it for brand voice and format preferences. If not, infer from the content itself.
+4. **Never silently render.** Every clip that ships must appear in the proposal the user approved.
+
+---
+
+## Inputs
+
+| Field | Required | Source |
+|-------|----------|--------|
+| Video path | Yes | First argument, or set via `set_video` |
+| Clip count | Optional | Second argument, e.g. `"5 clips"`. Default: propose what the content supports (3–8). |
+| Brief | Optional | Anything after the count, e.g. `"focus on the investor pitch moments"`. |
+
+---
+
+## Flow
+
+### Phase 1 — Inventory
+
+1. If a video path was given, call `set_video(file_path)` then `transcribe_podcast(file_path)`. If no path, read `get_ui_state` and use the current video.
+2. Read the packed transcript: `get_ui_state(include_transcript: true)`. This returns a compact phrase-grouped view with speakers, silence gaps, and energy peaks.
+3. If `.podcli/knowledge/` exists, read `01-brand-identity.md`, `02-voice-and-tone.md`, and `04-shorts-creation-guide.md` for show context. Skip silently if missing — `/auto` works on any content.
+4. Call `clip_history` to see what's already been shipped for this episode. Avoid duplicates in the proposal.
+
+### Phase 2 — Strategy Proposal (GATE)
+
+Emit a numbered strategy table. Do NOT render yet.
+
+```
+Proposed strategy for <video-label> (<duration>, <N> speakers):
+
+Inferred format: <talking-head | interview | montage | tutorial | travel | other>
+Inferred tone: <from voice fingerprint or knowledge/02>
+Target count: <N> clips (from arg, or inferred from content density)
+
+#1 [00:04:22-00:05:01] 39s S0 "<hook>"
+ Why: <one line — the angle, the stakes, or the moment>
+ Title: <suggested, ≤60 chars>
+ Style: <hormozi | karaoke | subtle | branded>
+
+#2 ...
+
+Unused peaks / skipped moments:
+- [00:18:45] — high energy but no standalone hook
+- [00:34:10] — great quote, but needs 40s of setup
+
+Confirm? (yes / redirect / change specific clips)
+```
+
+**Stop here. Wait for the user's response.** Do not call `batch_create_clips` on implicit approval — require an explicit "yes", "go", "ship it", or similar.
+
+### Phase 3 — Render
+
+Once the user confirms:
+
+1. Call `batch_create_clips(clip_numbers=[1,2,...])` with the confirmed list.
+2. Stream progress events back to the user as they arrive.
+3. When done, print the output paths and a one-line-per-clip summary.
+
+### Phase 4 — Persist
+
+Write a compact session log to `.podcli/sessions/<episode-slug>.md` (create the dir if missing):
+
+```markdown
+# <episode-label> — <ISO date>
+
+## Proposed
+<the strategy table>
+
+## Rendered
+- #1 → .podcli/output/<file>.mp4
+- #2 → ...
+
+## Skipped / redirected
+- <any user redirects>
+```
+
+This gives next week's session something to pick up from. Not required — skip silently on I/O failure.
+
+---
+
+## Error Handling
+
+- **Three-strike rule**: if any phase fails 3 times in a row, STOP and report. Do not continue with partial state.
+- **No transcription yet**: call `transcribe_podcast`. It runs Whisper + diarization + auto-packs — takes minutes on a 60-min episode.
+- **No packed view available**: `get_ui_state` falls back to raw segments automatically. Proceed, but flag lower reasoning quality in the proposal.
+- **User says "redirect"**: discard the current proposal, ask what to change, re-propose. Don't argue.
+
+---
+
+## What `/auto` Is NOT
+
+- Not a content-package generator — use `/produce-shorts` for titles/descriptions/thumbnails across the whole pipeline.
+- Not a self-eval loop yet — `timeline_view` (the visual composite that would verify cut boundaries on rendered output) is not built. For now, trust the cut planner and inspect output manually.
+- Not autonomous — the strategy gate is mandatory. No clips ship without explicit user confirmation.
+
+---
+
+## Completion
+
+Return one of:
+
+- **DONE** — Strategy approved, all N clips rendered, session log written.
+- **PARTIAL** — Some clips rendered, some failed. Report which succeeded and the failure reason for the rest.
+- **CANCELLED** — User redirected or rejected the strategy. No clips rendered.
+- **BLOCKED** — Upstream failure (transcription, no audio, corrupt source). Report the specific ask needed to unblock.
diff --git a/backend/cli.py b/backend/cli.py
index bed562e..746e8ed 100644
--- a/backend/cli.py
+++ b/backend/cli.py
@@ -2475,6 +2475,7 @@ def print_help():
print(f" {dim}${reset} podcli thumbnails \"Why AI Changes Everything\" --video ep42.mp4")
print()
print(f" {bold}PodStack{reset} {dim}(Claude Code slash commands):{reset}")
+ print(f" {accent}/auto{reset} One-verb pipeline: confirm strategy → render clips")
print(f" {accent}/prep-episode{reset} Full pipeline: transcript → publish-ready")
print(f" {accent}/process-transcript{reset} Extract clip-worthy moments from transcript")
print(f" {accent}/generate-titles{reset} Generate 8 title options with verification")
diff --git a/podcli b/podcli
index 30ad7e9..c45a41c 100755
--- a/podcli
+++ b/podcli
@@ -23,7 +23,7 @@ CMD="$1"
CMD_CLEAN="${CMD#/}"
CMD_CLEAN="${CMD_CLEAN#--}"
-PODSTACK_COMMANDS="prep-episode process-transcript generate-titles generate-descriptions plan-thumbnails review-content publish-checklist retro-episode plan-episode produce-shorts"
+PODSTACK_COMMANDS="auto prep-episode process-transcript generate-titles generate-descriptions plan-thumbnails review-content publish-checklist retro-episode plan-episode produce-shorts"
for pc in $PODSTACK_COMMANDS; do
if [ "$CMD_CLEAN" = "$pc" ]; then