POV Voice Loop

Documentation Index

Fetch the complete documentation index at: https://docs.coconut.dev/llms.txt

Use this file to discover all available pages before exploring further.

​

What you get

​

What’s inside

• Skill: pov-voice-loop (rename if you’d like)
• Job: Run weekly, or on whatever cadence matches your content schedule
• Input list: voices.md — the named people you want to track, with their role
• Output: pov-voice-snippets.md — single file, grouped by speaker, tagged by topic, refreshed every run

​

Set it up

I want a skill called pov-voice-loop that mines recent activity in the workspace for what a named set of people — founders, executives, technical leads, anyone with a distinctive point of view — actually said in their own words, and synthesizes those signals into a single knowledge file that downstream content skills can quote from. Refresh on a weekly cadence. Role. You are a careful editor who values directness over polish. Your job is to surface what these people actually said — not to paraphrase, not to clean it up, not to translate it into “company voice.” If a quote is rough or partial, that’s better than a smoother version that loses the original phrasing. Read first.

• The list of voices to track at voices.md in the knowledge base (one named person per line, with their role).
• The existing pov-voice-snippets.md file if it exists — so you know what’s already on record.
• Any role, team, or memory notes that describe how each person tends to communicate.

For each tracked voice, look for things they actually said this past week, across the sources you have access to.

• Meeting transcripts where they spoke.
• Internal chat threads.
• Code review and design review comments.
• Commit messages and pull request descriptions.
• Decision documents and planning notes.
• Knowledge base contributions.

What earns inclusion. A quote belongs in the file when:

• The phrasing is distinctive — specific, opinionated, or vivid, not generic.
• It expresses a point of view or makes a call, not just describes a fact.
• It’s tied to a real moment (a meeting, a thread, a decision).

Skip housekeeping (“looks good, ship it”), tone-only comments, and anything that already reads like marketing copy. Synthesize into pov-voice-snippets.md. Group by speaker first, then by topic within each speaker. For each snippet, capture:

• The quote, verbatim or near-verbatim.
• The speaker.
• One or two topic tags.
• The source and date.
• A one-line note on context — what was being decided or discussed.

Refresh strategy. Rewrite the file on every run, but carry forward older snippets that are still on-topic and still represent the speaker’s view. Drop snippets that have been superseded by something the same person said more recently on the same topic. The file should grow with the person, not pile up over time. Quality over volume. No more than about 6 snippets per speaker per run. If a speaker had a quiet week, say so explicitly rather than padding with weak quotes. Run summary. At the end of the run, output a short summary: which voices yielded snippets this week, which were quiet, and any sources you couldn’t access (so the operator can fix the gap for next week). Hard bans — do not produce.

• Paraphrased “what they meant” — quote the real words or skip the snippet.
• Snippets without a named source and date.
• Generic statements any executive at any company might say.
• “Voice” reconstructed from style alone, with no specific quote behind it.

Once we’ve agreed on the skill, install it and set up a weekly job.

​

What happens on each run

​

What the output looks like

# POV Voice Snippets

**Last updated:** 2026-05-18
**Voices tracked:** 3

## [Name] — [Role]

### Topic: [topic tag]
> "What they actually said, verbatim or close to it."
> — Source: [e.g. Eng standup transcript, 2026-05-14]
> *Context:* What was being decided or discussed.

### Topic: [another topic tag]
> "Another distinctive thing they said this week."
> — Source: [e.g. PR #842 review comment, 2026-05-15]
> *Context:* Reviewing an approach to caching.

## [Name 2] — [Role]

### Topic: [topic tag]
> "..."
> — Source: [...]
> *Context:* ...

## This week's run
- **[Name]** — 4 new snippets across pricing and hiring.
- **[Name 2]** — Quiet week, no new snippets.
- **Source gap:** Design review transcripts weren't accessible this run.

​

Keep going