Project Update

project-update — in-project update engine (Tier 1)

You are inside one project repo and need either to (a) catch up on its current state, or (b) build its canonical dated update bundle for a weekly checkpoint or a PI 1:1. This skill wraps a deterministic, stdlib-only engine that mines only this one repo — git history, narrative docs, evidence files — and synthesizes the result, driven by the project's own .sync/manifest.yaml.

Isolation invariant. This engine has NO dependency on LabSync. A project + this skill is sufficient to build that project's own update. LabSync (Tier 2) later reads the bundle this produces and runs the post-meeting writeback — it does not mine the project itself. See 36-LabSync/docs/ADR-001-two-tier-update-architecture.md.

The engine lives next to this file at engine/project_update/ (a peer of the compute-strategy / compute-validation / campaign-orchestration skills).

When to use this

• An agent starts work in a project and asks "what's the current state?" → orient.
• Sean (or LabSync's dispatch) says "build the weekly update" / "prep the Friday 1:1" / "aggregate the relevant files into the update dir" → build.
• You need the deterministic, no-LLM floor for producing a meeting bundle.

Do not use this to author figures, edit the manuscript, run simulations, or write into another project. It mines + synthesizes; it only triggers the project's own deck builder, never authors slides.

Prerequisite: the manifest

The engine reads <repo>/.sync/manifest.yaml. If it is absent, the engine errors with a hint. Create one per 36-LabSync/docs/MANIFEST_SCHEMA.md — declare the docs the project actually has (status/tracker/changelog/agents, omit what is missing), tracker_format.pi_flag_markers (the markers that escalate a WORKSTREAM row to a PI ask, e.g. ["🔴", "flag to hendrik", "needs pi confirm", "sign-off"]), evidence_globs, the deck builder, and bundle.root (updates/pi-meetings). Absent docs degrade gracefully — no changelog → no window-news; no tracker → no Ask-PI/drift; no deck → skip deck.

The two modes

1 · orient — get me up to speed (ephemeral, not committed)

python3 /path/to/skills/project-update/engine -m project_update \
  orient --repo "$(pwd)" --since 2026-05-22
# or, if the engine package dir is on PYTHONPATH:
PYTHONPATH=.../skills/project-update/engine python3 -m project_update orient --since 2026-05-22

Prints a current-state brief to stdout: scientific focus (from STATUS), what moved since --since (CHANGELOG news or ranked commits), open PI asks (🔴 workstream asks + the D.N decision queue), and blockers. It is transient — not written to a committed file. Use it to catch up before doing work.

2 · build — write the canonical dated bundle

PYTHONPATH=.../skills/project-update/engine python3 -m project_update \
  build --repo "$(pwd)" --date 2026-05-29 --since 2026-05-22 --kind pi-meeting
# add --no-deck to skip the deck rebuild; --kind weekly for a no-meeting checkpoint

Writes <repo>/<bundle.root>/<date>/ per 36-LabSync/docs/BUNDLE_CONTRACT.md:

• meeting.yaml — machine-readable parse anchor (provenance incl. repo HEAD, generated_by: project-update@<version>, the window, the artifact list).
• synthesis.md — full reconciliation: scientific focus, this-week's-developments (CHANGELOG), milestones (doc-only, kept OUT of drift), activity ledger, claimed-vs-real drift, open PI-decision queue (🔴 asks lead).
• PREP.md — in-the-room script: what to open, 90-sec opening, headline, walk- through, Ask PI (🔴 workstream asks first, then D.N), engineering FYI, drift.
• artifacts/ — symlinks (never copies) to the regenerated deck + manuscript / SI PDFs, always resolving to the current build.

The deck step triggers the manifest's deck.builder. If python-pptx is missing or the builder fails, it degrades with a clear message and still symlinks the existing built deck.

Boundaries (read before you run build)

• This repo only. The engine never reaches into another project; cross-project knowledge lives in LabSync's registry, never in a manifest.
• Additive. A build writes a new dated dir. Re-running for the same date regenerates only the engine-owned files (meeting.yaml, PREP.md, synthesis.md, artifacts/) and never touches human/LabSync files in that dir (TRANSCRIPT.md, feedback-extracted.md, action-items.md, writeback-log.md).
• Never authors figures. It triggers the project's deck rebuild; the figure source stays owned by the project.
• Git hygiene. The bundle is purely additive — commit the new dated dir as its own commit. Do not stage unrelated pending work from other agents.
• Post-meeting is LabSync's job. Sean pastes TRANSCRIPT.md; LabSync extracts directives and writes them back into the TRACKER. The engine stops at prepped.

Quick checklist

• Does <repo>/.sync/manifest.yaml exist + declare the docs this repo has?
• Did I pick the window (--since = last checkpoint, --date = meeting date)?
• For a build: is --kind right (pi-meeting vs weekly)?
• After build: does the synthesis lead with the 🔴 PI-flagged ask + CHANGELOG news, and do artifacts/ symlinks resolve to real files?
• Am I committing ONLY the new dated dir (+ a new manifest if I created one)?