| name | up-apply |
| description | Apply a follow-up conversation to patch or steer existing docs |
Apply a follow-up conversation
Patch existing docs based on a new research conversation. This is NOT a full bootstrap — it updates what changed and leaves the rest alone.
Input
$ARGUMENTS can be a path to a conversation file. If empty, check for conversation.md at the repo root. If that doesn't exist, ask the user.
The conversation can be anything — a ChatGPT/Claude export, raw copy-paste, meeting notes, a markdown brain dump. If it wasn't started with the snapshot context prompt, work harder to map changes to existing docs and flag anything ambiguous.
How it works
Read the conversation, read the current project state (index, specs, phases, memory, workflow), and figure out what changed.
For long, complex conversations (2000+ lines with multiple pivots), use the Large Conversation Protocol from /up-bootstrap — scan and index first, confirm with the user, then extract from confirmed sections only.
What to patch
- Specs — update affected files only, don't rewrite unchanged specs. Mark changed sections with the conversation source.
- Phases — material scope changes get a
kind: steering phase (per AGENTS.md). New features get new delivery phases. Never modify completed phases.
- Project memory — append new decisions, never overwrite. If a decision is superseded, mark the old one and add the new one.
- Index — update
.unpack/docs/index.md with any new or changed phases.
What NOT to patch
- Specs the conversation didn't touch
- Completed phases
- Existing decisions (append, don't overwrite)
Archive and report
Ask for a conversation name (e.g., "auth-iteration"), move the file to .unpack/conversations/NNN-name.md, and report what changed — specs updated, phases added, decisions captured, contradictions found, and what's next.
If the conversation contradicts existing specs, flag it and ask the user — don't silently overwrite.