Add podtx summarize for episode summaries and takeaways

[frarredondo]

Summary

Raw transcripts are long. Users often want a short summary, key points, and timestamped quotes without re-listening.

Motivation / evidence

Library already contains hour-long interviews (12k–16k+ words). Manual summarization per episode doesn’t scale.

Proposal

podtx summarize path/to/episode.json
podtx summarize --feed corecursive-coding-stories --limit 5

Produce a sidecar summary (e.g. .summary.md or JSON) with:

• short overview
• key points / takeaways
• optional notable quotes with timestamps

Use a pluggable backend (local model and/or API), configured explicitly—no silent external calls.

Out of scope

Full chat-with-podcast UI, automatic publishing to blogs/newsletters.

Acceptance criteria

• Can summarize from existing transcript JSON without re-running ASR
• Output format is stable and documented
• Backend/credentials behavior is explicit and offline-friendly by default where possible
• Docs / --help updated

[frarredondo]

This was generated by AI during triage.

Agent Brief

Category: enhancement
Summary: Add offline-friendly episode summarization (summary + takeaways + timestamped quotes)

Current behavior:
The library stores full interview transcripts (often 12k–16k+ words) as .json+.txt plus timed segments, and a reformat path already rebuilds outputs without re-running ASR. There is no summarization command; users must read full text or summarize manually per episode.

Desired behavior:
A new podtx summarize command generates a stable sidecar summary from existing transcript JSON without re-running ASR. It produces a short overview, key points/takeaways, and optionally notable timestamped quotes. The summarization backend is pluggable and explicitly configured — no silent external network calls — and defaults to an offline-friendly option where possible. Output format is stable and documented, and single-episode and batch-by-feed modes are supported.

Key interfaces:

• CLI: podtx summarize <path/to/episode.json> and podtx summarize --feed <slug> [--limit N] (or equivalent batch flags); credentials/backend selection is explicit via flags/config/env, never implicit
• Input: existing transcript JSON (text + segments) as the source of truth; no audio re-ingest
• Output: a sidecar file per episode (e.g. .summary.md or structured JSON) with stable fields for overview, key points, and optional quotes with timestamps; format is versioned/documented

Acceptance criteria:

• Can summarize from an existing transcript JSON without invoking the ASR engine
• Supports both single-file (path/to/episode.json) and batch (--feed <slug> [--limit N]) invocation
• Output is a stable, documented sidecar format containing at least overview + key points, with optional notable quotes carrying timestamps
• Backend/credential behavior is explicit: the user must opt into any external API use; default path is offline-friendly or clearly requires local model setup, with no silent network calls
• Docs: README and --help describe invocation, output format, and backend configuration; examples cover single and batch modes

Out of scope:

• Full chat-with-podcast UI or conversational agent
• Automatic publishing to blogs/newsletters or third-party platforms
• Semantic search over summaries (search is a separate feature)
• Re-transcription or audio re-processing during summarization

[frarredondo]

Shipped by PR #33 (commit df7409b). Verified against this issue's acceptance criteria:

• Summarize from existing transcript JSON without re-running ASR: met (podtx summarize <path>.json)
• Batch modes: met (--feed <slug> [--limit N], plus --all)
• Stable sidecar format (overview, key points, optional timestamped quotes): met (.summary.json / .summary.md)
• Backend/credentials explicit and offline-friendly: met (default fake extractive, no network; openrouter/opencode/lmstudio opt-in; keys via podtx auth set, env, or --api-key)
• Docs: --help met; README still missing, tracked in Document podtx summarize in README #37

Closing as done; the README gap is tracked separately in #37.