post-angles

post-angles

Answers the question a full backlog does not: what is worth a post this week, and to whom.

Having material is not the same as knowing what to say about it. Ten typed items and thirty commits do not produce a post, because the angle is a claim about a reader and the record only holds claims about the work. This is what stops people building in public: sitting down with a complete record, finding nothing that feels postable, and closing the tab. The material was there. The angle was missing.

The opposite failure matters as much and is the one a model will commit automatically. Asked for post ideas, it will always produce four — including in a week where nothing happened. A tool that cannot say nothing here yet is a tool for posting about nothing, and it poisons every other output in the family, because after the third invented angle nobody trusts the sourced ones either.

Three rules govern this skill:

• It can decline, and declining is a designed output, not an error path.
• It proposes and never ranks. No angle is presented as the best one, and none is scored. The user picks.
• It never estimates performance. No predicted reach, no "this will do well", no best-time-to-post. There is no evidence for any of it, and inventing it here would contaminate a skill whose whole value is that its claims are checkable.

Instructions

Step 0 — Detect your surface

Using Bash availability:

• Claude Code — Bash works and there's a real filesystem. Full flow.
• Claude.ai — no filesystem. Work from the conversation, and say plainly that you cannot check against prior posts, so a repeat is possible. Never claim a record you cannot read. Don't error.

Step 1 — Resolve the sources

1. The backlog — posts/material/, resolved the way post-export resolves it: a CLAUDE.md declared path, then an existing folder, then canon.
2. The record — changelog/, docs/decisions/, the explorations tier, handoff/.
3. What has already been said — posts/*.md.