Content Idea Capture

Capture a content idea

Catch a spark before it evaporates. Seconds, not a conversation.

Format: _shared/contracts/ideas-inbox-format.md. It is the source of truth — where this file and the contract disagree, this file is the defect.

Capture

One exchange. Take the author's words, append an entry to content-system/ideas.md, confirm in one line, stop.

• Record the idea in the author's own words. Do not sharpen it, expand it, retitle it, or make it sound better. A capture that comes back polished is a capture the author has to check.
• Scope is optional. Offer the scopes that have a strategy file — walk content-system/strategies/ — and accept "none" instantly. Never block on it. Most sparks arrive before anyone knows where they belong.
• ID is YYYY-MM-DD-<letter>, the letter incrementing within the day. It exists so the author can say "the second one" while the idea waits.
• Several ideas in one message is one capture, not several. Append them all in one write, letters incrementing across the batch, and confirm in one line. Turning a batch into five exchanges is the same failure as interviewing — it makes the fastest action in the system the slowest.
• A scope the author has just named may be applied to the whole batch, provided you say so. "All five under areaOfFocus/product-craft — tell me if any of those belong elsewhere." An announced scope is corrected in seconds; a scope question per row is the interview this skill exists to avoid. Say nothing and it is an inference, which is a different thing entirely.
• If the author volunteers why an idea matters, record it — in the same write, verbatim. It goes in the entry's own body, under its heading. This is the one part of a capture that can run to a paragraph, and it is the part that decays fastest: a line that names a topic and promises a payoff reads fine today and is unrecoverable in a week.

Context is accepted, never requested

Never ask for it. Not "would you like to add why?", not a follow-up after the confirmation, not a nudge on the entries that look thin. Asking on every capture costs the ten-second promise for the minority of ideas that carry any, and it is the interview under another name.

Never write it yourself. If the author did not say why it matters, the entry has no body. An inferred reason is a fabricated one, and it is worse than nothing here because it will later be read as the author's own thinking.

An entry with no context is complete, not unfinished - a heading and a meta line is the ordinary case. Never render it as missing, never list which ideas lack one.

The one hard rule

Capture never interviews.

The refinement gates protect the moment an idea becomes a piece. They do not belong at the moment it becomes a row. An interview here makes the fastest action in the system the slowest — and then the author stops using it and the idea is lost entirely, which is the failure this skill exists to prevent.

No angle. No "what's the most important point?". No audience. No persona. Those are Phase 1 of content-post-writer, and they are that skill's job.

What's waiting

Read content-system/ideas.md and render a compact list - ID, idea, scope - one line each. The count is the number of entries - no filtering, no status to interpret. That is the whole reason the file holds waiting items only.

That list is a view, never written back. The file holds the entries; a stored summary would be a second representation of the same ideas, and two representations drift. Show context only when asked for a specific idea, or the compact list stops being compact.

Promotion — an idea that is becoming a piece

Promotion is for an idea you have started turning into something: an angle forming, an outline, notes that are about the writing rather than about the idea. It is not the place to record why an idea matters — that is ## Context, and it does not move the idea out of the inbox.

Remove the entry from the inbox and create a bundle folder:

content-system/posts/YYYY-MM-DD-slug/backbone.md

with frontmatter carrying from-idea: (the ID and the original text) and scope: if known, then ## Idea — the capture, verbatim — and ## Notes.

If the entry has context, it becomes the opening ## Notes. It travels with the idea and leaves ideas.md with it - copying it would leave two versions to disagree. With no context, ## Notes starts empty as before.

No angle is required. No plan. No commitment to finish. This is the point of promotion: an idea being actively worked is neither waiting nor written, and without a place to sit it would have to stay in the inbox pretending to be untouched. The bundle folder is where refinement happens — notes now, angle when it settles, and the rest as content-post-writer fills it in.

A promoted idea may be abandoned. The folder stays where it is, counted as in-progress. Nothing drags it back to the inbox, and nothing needs cleaning up.

Drafting straight from an inbox idea is a promotion — it creates the folder and removes the row in one step. Hand off to content-post-writer from there.

Never route to the assessment pipeline

.specify/assessments/ is a build/don't-build gate for capabilities. It asks for goals, non-goals, success metrics, appetite and solution options, and its verdict hands off to /speckit-specify. Those are questions about shipping software, not writing a piece — and it sits outside content-system/, so anything routed there breaks the moment this folder is copied to another context.

An idea for this system belongs in assess. An idea for a piece belongs here. A content idea that needs real work gets promoted and refined in its own backbone; it does not need a heavier gate.

Dropping

Move the idea to the ## Dropped table with the date and a one-line reason. Its context, if any, goes with it and may be pruned. Ideas are allowed to die — an inbox that only accumulates stops being an inbox and becomes a guilt pile.

That section is prunable at any time and nothing reads it.

Degradation and cold start

Works with nothing else present — no strategy, no plan, no posts, no prior ideas. A missing content-system/ideas.md is created on first capture, never scaffolded ahead of time — header, table header and an empty ## Dropped section, written in the same turn as the first row. The cold start is not a special case and needs no announcement beyond the usual one-line confirmation.

Verify

After appending, moving, or removing an entry, re-read the file and confirm it still parses — one ### heading per waiting idea, each with a Captured meta line, nothing left in two places, and the ## Dropped table intact. Correct once; on a second failure say so plainly. A malformed table silently miscounts the pipeline's front stage, which is the one thing this file exists to make countable.

What this must never become

• An interview.
• A tagging, priority, or scoring system. ## Context holds the author's reason, never a rating of it.
• A format that asks for context, marks an entry without it as incomplete, or reports which ideas lack one.
• A stored summary table of waiting ideas. Render one on request; writing it into the file makes a second copy that drifts.
• A place where ideas get developed rather than parked.
• A second inbox alongside the outer workspace's _editorial/_ideas/ — this one reads nothing outside content-system/.
• A router into .specify/assessments/.