/mstack:paper-status

Stage: any · Voice: project-manager

Opening a session, returning after a gap, "where were we?", "what's next?". Read-only apart from an offered paper.status correction.

Procedure

1. Inventory in one call:

   python3 "${CLAUDE_PLUGIN_ROOT}/bin/mstack-status"

   It finds .mstack/ at or above the current directory (exit 1 and a pointer to /mstack:mstack-init if there is none), reads .mstack/config.yaml, and prints: every pipeline artifact by stage with its state (present / thin / stub / missing / optional), date, and verdict line, where thin means the file exists but is too small or too repetitive to be a real artifact of its kind; verdicts that are stale because an input changed after they were written (a results audit older than code/02-analyze.R, an identification review older than methods.tex, a lit map older than lit/index.md, PDFs dropped in lit/pdf/ but never ingested); the stage the artifacts imply; whether that matches paper.status; and the required artifacts still missing up to that stage. Do not re-derive any of this with your own globs and greps.

2. Spot-check what the verdict rests on. Open every artifact the script marks thin, and the one present artifact that anchors the derived stage (the furthest one). If any of them is placeholder text, say so and derive the stage yourself from the last real artifact; the script's heuristics are size and repetition, and your reading beats them.
3. Reconcile paper.status. If the script's derived stage, or your corrected one, disagrees with config, say so and offer to update the field in .mstack/config.yaml; never change it unasked.
4. Recommend exactly one next step: the earliest required gap wins over a shinier later stage, and a stale verdict counts as a gap. Name the skill and the reason in one sentence.

Outputs

A printed report, in this order, and nothing written to the paper folder:

1. The config line as the script prints it: title, paper.status, format, design, prereg, target journals.
2. Bottom line, four short sentences (under 80 words): the stage and whether it matches paper.status; which verdicts are stale (an input changed after they were written; a placeholder is a gap, not a stale verdict); the one next skill and why.
3. One per-stage table, the script's rows with your spot-check corrections already applied (state, date, verdict or a short note). One table is the inventory; do not paste raw script output and then re-summarize it.
4. Only what changes the next step: a placeholder you found behind a present or thin label, a mismatch to correct. Observations that do not change the recommendation stay out.

Write for the author, not about the process: no "the script says", no state-versus-correction columns, no mention of which heuristic decided what. Report the corrected state and, where you overrode the script, one short note in the row.

With consent only: paper.status corrected in .mstack/config.yaml.

Anti-patterns

• Guessing the stage from the conversation. The artifacts on disk are the record, and the script reads them.
• Three next steps. One; the pipeline is ordered for a reason.
• A stub counted as done. The script strips template comments before deciding; trust its stub state, open anything it calls thin, and spot-check the artifact the stage rests on.
• A report longer than the paper's problems. Bottom line, one table, then only what changes the next step.