Add speak-summary skill: turn any text or summary into a listenable MP3

[Sam-Rowe]

What this adds

A speak-summary skill that converts text, markdown, or the output of another skill into an MP3 using local, CPU-only neural text-to-speech.

Why

There is no audio generation skill in the collection today. I checked before building: resemble-detect does deepfake audio detection (the opposite direction), and the media skills (generate-image, screen-recording, image-manipulation-image-magick) are image and GIF only — screen-recording explicitly notes "No audio support in GIF". So this fills a gap rather than duplicating existing coverage.

The use case is asynchronous catch-up: generate a digest, listen to it while commuting, cooking, or away from a screen.

Designed to chain

This is deliberately a terminal step, not a self-contained tool. It does not gather or summarise anything — that would duplicate roundup, which already does multi-source date-ranged briefings well. The intended pattern is:

• roundup → speak-summary — spoken status briefing
• daily-prep → speak-summary — tomorrow's schedule, listened to tonight
• meeting-minutes → speak-summary — catch up on a meeting you missed

The upstream skill owns what to say; this one owns how it sounds. Pairing it with a scheduled workflow gives you a digest waiting before breakfast.

The part that carries the value

Most of SKILL.md is guidance on rewriting written prose for the ear before synthesising. Piping markdown straight into a TTS engine gives you something technically correct and genuinely unpleasant to listen to — read-aloud URLs, bullet fragments with no connective tissue, 2026-08-20 spoken as digits.

So the skill instructs the model to expand abbreviations on first use, speak dates naturally, replace bullets with connective prose, signpost transitions, drop visual-only content like tables and code blocks, and close with the actions since a listener cannot scroll back.

Privacy and portability

Synthesis uses Kyutai pocket-tts, a small neural model built to run on CPUs. This was a deliberate choice over a cloud TTS API:

• Nothing leaves the machine, which matters when the summarised content is confidential.
• No API key, so there is no paid-service dependency.
• CPU-only, so it runs unattended in a headless cloud agent or CI container just as well as on a laptop.

It is roughly 6x faster than real-time; a 5-minute briefing renders in about 50 seconds.

Implementation notes

scripts/tts.sh bootstraps pocket-tts into a cached virtualenv on first use, so there is no manual setup step.

Two things it handles that are easy to get wrong:

• Python version. pocket-tts requires >=3.10,<3.15. The script searches for a compatible interpreter rather than assuming python3 qualifies, and fails with actionable guidance when none exists. I hit this genuinely while testing on Python 3.15, where the naive approach fails confusingly mid-run.
• Chunking. Quality degrades on long inputs, so text is split on sentence boundaries into ~600 character chunks (falling back to comma boundaries for very long sentences), synthesised individually, and concatenated. Output is mono 96kbps MP3 — small enough to sync to a phone, fine for speech.

ffmpeg is preferred for encoding, with a fallback to macOS afconvert (emitting .m4a) when it is absent.

Testing

• npm run skill:validate — passes (412 skills valid)
• npm run build — regenerated docs/README.skills.md
• Synthesised multi-chunk briefings end to end and verified durations against word counts (~180 wpm) to confirm no text is silently dropped
• Verified the clean-slate bootstrap with pocket-tts absent from PATH: creates the venv, installs, and synthesises successfully
• Verified the failure path on an unsupported Python produces a clear message rather than a crash

Note on voice cloning

pocket-tts supports voice cloning, so the skill documents it — along with explicit guidance not to clone a real person's voice without their consent.

[github-actions]

🔒 PR Risk Scan Results

Scanned 2 changed file(s).

Severity	Count
🔴 High	0
🟠 Medium	0
ℹ️ Info	1

Severity	Rule	File	Line	Match
ℹ️	skill-script-touched	skills/speak-summary/scripts/tts.sh	1	skills/speak-summary/scripts/tts.sh

This is an automated soft-gate report. Findings indicate review targets and do not block merge by themselves.

[github-actions]

🔍 Vally Lint Results

✅ All checks passed

Scope	Checked
Skills	1
Agents	0
Total	1

Severity	Count
❌ Errors	0
⚠️ Warnings	0
ℹ️ Advisories	0

Summary

Level	Finding
ℹ️	✅ speak-summary (2/2 checks passed)
ℹ️	✓ [spec-compliance] All 1 skill(s) are spec-compliant.
ℹ️	✓ spec-compliance: All spec checks passed.
ℹ️	✓ [valid-refs] All file references across 1 skill(s) are valid.
ℹ️	✓ valid-refs: All file references resolve to existing files within the skill directory.
ℹ️	1 skill(s) linted, 1 passed

Full linter output

### Linting skills/speak-summary
✅ speak-summary (2/2 checks passed)
    ✓ [spec-compliance] All 1 skill(s) are spec-compliant.
        ✓ spec-compliance: All spec checks passed.
    ✓ [valid-refs] All file references across 1 skill(s) are valid.
        ✓ valid-refs: All file references resolve to existing files within the skill directory.

1 skill(s) linted, 1 passed

[Copilot]

Pull request overview

Adds a terminal skill that converts written summaries into locally generated audio.

Changes:

• Adds guidance for rewriting text for listening.
• Adds local pocket-tts synthesis, chunking, and encoding.
• Registers the skill in generated documentation.

Show a summary per file

File	Description
skills/speak-summary/SKILL.md	Defines usage, preparation, voices, and output guidance.
skills/speak-summary/scripts/tts.sh	Bootstraps TTS and generates MP3/M4A audio.
docs/README.skills.md	Adds the skill to the catalog.

Review details

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 3/3 changed files
• Comments generated: 2
• Review effort level: Balanced