bip-ms-tuckin

name	bip-ms-tuckin
description	Persist manuscript session state before context reset

/bip-ms-tuckin

Flush manuscript session state to durable storage before a context reset or session end. Run this when context is getting long or before stopping.

Usage

/bip-ms-tuckin

Workflow

Step 1: Check for uncommitted manuscript changes

cd <manuscript-root>
git diff --stat
git diff --cached --stat

If there are uncommitted changes to TeX files, ask the user whether to commit them. Draft a commit message summarizing what changed (e.g., "Update implementation notes: Phase 1b done, branchScale normalization"). Do not commit without confirmation.

If there are only untracked ISSUE-*.md files, note them but do not commit (these are intentionally untracked per CLAUDE.md).

Step 2: Push EPIC body updates

For each tracked repo in .ms-config.json, check whether the EPIC body was edited this session (compare the updatedAt from the start of the session to the current updatedAt). If not pushed yet, push any pending EPIC body updates:

gh issue view <epic-number> --repo <org/repo> --json updatedAt

Report which EPICs were pushed and which were already up to date.

Step 3: Update .ms-config.json

Check whether fetch_cmds should be updated based on new experiments or results that arrived this session. If new result paths were discovered but not added to fetch_cmds, note them for the user.

Step 4: Update memory

Update the auto-memory files to reflect the session's work:

1. project_dasmfit_status.md (or equivalent project memory):

   • Mark newly completed items
   • Update open issue list
   • Note the current bottleneck
   • Record any key decisions made this session
   • Record in-flight research threads: analyses requested on PRs/issues this session, open scientific questions, and what each thread is waiting on — so the next session resumes the orchestration, not just the manuscript. (A thread that has completed is what triggers a paper update; one still open stays here.)

2. project_pending_decisions.md (if it exists):

   • Remove resolved decisions
   • Update status of pending decisions
   • Add any new unresolved items

3. Feedback memories: If the user corrected an approach or confirmed a non-obvious choice, save it.

Only update memories that changed. Do not rewrite unchanged files.

Step 5: Verify build

make pdf

Ensure the manuscript builds cleanly. If it doesn't, fix the build error before completing tuckin.

Step 6: Report

Print a summary:

## Tuckin Complete

### Manuscript
- Committed: <yes/no, commit hash if yes>
- Build: <clean/broken>
- Uncommitted ISSUE files: <count>

### EPICs
- <repo> i<N>: <pushed/up-to-date/skipped>

### Issues filed this session
- <repo>#<N>: <title> — <status>

### Pending for next session
- <brief list of what's unfinished>

### Key decisions
- <any decisions the next session should know about>

Safe to reset context.