Idea — fast capture of a single idea

Part of the Personal Corp framework — running a one-person business through AI agents.

One pipeline: your voice → a provenance-tracked folder in your ideas repo → (optional) an issue mirror on a GitHub Project view. The folder is the source of truth; everything else is a pointer to it.

This skill is self-contained — it inlines the folder structure, status enum, and provenance formats below. You don't need any external canon file.

What this skill is NOT for

• Mining ideas out of a transcript / meeting / thread — that's bulk extraction from a source. Use a separate extraction skill for that. This skill captures ONE idea the user is voicing right now.
• A dated work task with a deadline — use an issue tracker / the manager skill. Ideas have no deadline; they're seeds.

Setup

Before first use, define this in your project's CLAUDE.md:

## Idea Config

### Ideas repo path
Where idea folders live (one folder per idea):
- ideas_repo: ~/Projects/ideas

### Default author
Name written into `first_voiced_by` when the user doesn't say otherwise:
- default_author: Me

### GitHub mirror (optional)
Mirror each idea as an issue (added to a Project view filtered by `label:idea`):
- mirror_enabled: false
- owner: your-github-handle               # only if mirror_enabled: true
- ideas_repo_name: ideas                  # GitHub repo name for the issue, only if mirror_enabled: true
- project_number: <your Project number>   # only if mirror_enabled: true

(if mirror_enabled is false, the skill only writes to the ideas repo — no issue, no Project)

No separate init skill needed — this section is the setup. If mirror_enabled is false, the GitHub Project step is skipped entirely; the idea still lives as a folder.

One question — max

Ask at most one short question, and only if the idea's essence can't be extracted from the wording (what is it / for whom). Do NOT ask about mechanics, pricing, risks, or tech stack — capture the raw material as-is and write the gaps into next-steps.md as open questions. The question "how should I record it — issue / file / both?" is forbidden: the form is fixed by this skill. Never block on form.

Status enum (in idea.md frontmatter)

• seedling — just voiced, not yet worked on
• exploring — first research gathered, market.md filled
• building — landing / pilot / prototype in progress
• live — a public artifact exists
• parked — set aside, reason in next-steps.md
• dropped — closed, reason in next-steps.md

A freshly captured idea is always seedling.

Provenance formats (rows in sources.md)

Every idea needs at least one source. Each sighting is one row. Genericize the source kind — no real internal paths:

Steps

1. date — resolve the real date. Between 00:00–04:00, "today" means the previous calendar day.
2. Dedup against the index. Read INDEX.md at the root of your ideas repo. Match is semantic, not string ("a board of advisors run on agents" ≈ advisory-agent-grounded-data). If unsure, open the candidate's idea.md to compare.
   • Match found → this is a repeat sighting, not a new idea:
     • append a row to the existing folder's sources.md,
     • in idea.md bump last_touch and increment sources_count (do NOT touch first_voiced),
     • if mirror is enabled, add a one-line comment to the issue mirror ("voiced again: {date}, {context}"),
     • regenerate INDEX.md, commit, report. STOP. Do not create a new folder.
3. New idea → create the folder ideas/YYYY-MM-DD-{slug}/ with the template files below.
   • Slug: kebab-case, 3–5 words, by meaning not by form (live-sales-coach-b2b, not b2b-product-1). The date prefix is the date of first capture — it never changes on later sightings.
   • In idea.md: title, status: seedling, first_voiced_by: {default_author from config unless user said otherwise}, an "Essence" of 1–2 sentences, and the raw verbatim wording preserved exactly as voiced. Interpretation, if any, goes in a separate section — never overwrite the raw wording.
   • In sources.md: one row for this sighting (see provenance formats).
   • market.md is optional (leave the stub; fill only when researched).
   • next-steps.md holds open questions and the "what next?" — this is where pricing/mechanics/stack gaps land.
4. Regenerate INDEX.md (format below). It's auto-generated — don't hand-edit individual rows; rewrite the whole file from the folders.
5. GitHub Project mirror — OPTIONAL (only if mirror_enabled: true in config). The folder is the source of truth; the issue is just a pointer for board view. See "Project mirror" below.
6. Commit + push to the ideas repo: feat: idea {slug} (invoking this skill is consent to record; the ideas repo is private).
7. Report — one line per item: folder path, issue #N (if mirrored), and the Project view link (if mirrored).

Template files

Create these four files in ideas/YYYY-MM-DD-{slug}/. Keep them minimal — the folder is a seed, not a spec.

idea.md

---
title: {short human title}
status: seedling
first_voiced_by: {default_author}
first_voiced: {YYYY-MM-DD}
last_touch: {YYYY-MM-DD}
sources_count: 1
---

# {title}

## Essence
{1–2 sentences — what it is and for whom.}

## Raw wording
> {the idea exactly as voiced — verbatim, do not paraphrase}

## Interpretation (optional)
{any reading/expansion goes here — never inside Raw wording}

sources.md

# Sources

| Date | Who | Context | Reference |
|---|---|---|---|
| {YYYY-MM-DD} | {default_author} | first voiced | direct /idea invocation, session, {YYYY-MM-DD} |

market.md (optional stub)

# Market

_Not researched yet. Fill when status moves to `exploring`: analogues, demand signals, where it could fit._

next-steps.md

# Next steps

## Open questions
- {gaps you didn't ask about — mechanics, pricing, audience, stack}

## What next
- {landing? brainstorm? park? close? — leave blank if undecided}

INDEX.md regeneration

INDEX.md is auto-generated — never hand-edit. Rewrite it whole from the current folders. Structure:

# Ideas index

Updated: {YYYY-MM-DD}

## Active ideas
| Slug | Status | First voiced | Sources count | Last touch |
|---|---|---|---|---|
| live-sales-coach-b2b | seedling | 2026-05-01 | 1 | 2026-05-01 |

## Parked / dropped
| Slug | Status | First voiced | Last touch |
|---|---|---|---|

## By status
- seedling: N
- exploring: N
- building: N
- live: N
- parked: N
- dropped: N

Project mirror (optional — only if mirror_enabled: true)

Each idea folder gets one issue mirror in the ideas repo, added to a GitHub Project view filtered by label:idea. The folder is the source of truth; the issue is a pointer for quick board scanning. Don't copy folder contents into the issue body.

Mirror format:

• title: idea: {title}
• label: idea (the only label — don't invent others)
• body: link to ideas/{folder}/idea.md + one "Essence" paragraph + status + one provenance line. Nothing more.

Commands (after creating a new idea folder):

url=$(gh issue create -R {owner}/{ideas_repo_name} \
  --title "idea: {title}" \
  --body "{body — link + essence + status + provenance}" \
  --label idea)

# Add to the Project board (number from config), so it appears in the label:idea view
gh project item-add {project_number} --owner {owner} --url "$url"

On rename → update the issue title. On dropped → close the issue with a comment stating the reason. On a repeat sighting (step 2) → just add a one-line comment, no new issue.

If mirror_enabled is false, skip this whole section — the idea still lives correctly as a folder.

Failure without the skill → rule

Red flags — stop

• About to create an issue without a folder → step 3 was skipped.
• Writing a body longer than ~10 lines → you're copying, not linking.
• Asked the user a second question → just capture it as-is.
• Finished without item-add (when mirror enabled) or without commit → steps 5–6 not done.

Cross-references

• Mining ideas out of a transcript / meeting / thread — a separate extraction skill (bulk source → many ideas). This skill is single-idea capture only.
• A dated work task with a deadline — the manager skill / your issue tracker, not an idea folder.