Skip to main content

Introduction

PM Workspace Kit is an opinionated template for the kind of work that happens before production code: product discovery, migration planning, cross-team reviews, and keeping all of it traceable after the initial excitement wears off.

It was extracted from a real 18-month ERP migration planning effort. The domain content (product entities, module playbooks, specific SLOs) was stripped out. What's left is the structural scaffolding that worked — the scripts that enforce it in CI so it doesn't rot, plus a CLI (pmk) and a Slack gateway that turn each PM verb into a repeatable conversation against your real codebase.

The product, one line​

PMK turns PM documents, code intelligence, and human expert answers into a traceable knowledge loop for code-aware teams.

Three product surfaces​

The kit has three surfaces. The first two carry the main value today; the third is an early shell that the project will return to once the gateway loop is mature.

1. Local doc/code workflow — pmk CLI + templates + traceability​

The day-to-day workhorse. PM verbs as named conversations (pmk propose, pmk ask, pmk case, …), backed by:

  • traceability.js — markdown front-matter validation, Mermaid dependency graph, reverse-lookup, orphan report. Runs in CI so links don't rot.
  • confluence-sync.js — pulls reviewer comments and status labels back from Confluence so wiki activity doesn't evaporate.
  • ADR template (MADR-style) with slots for Alternatives Considered and Consequences.
  • Module migration playbook with a Stage 0–4 Strangler Fig checklist and quantitative exit criteria.
  • Handoff kit — PR / Code review / Runbook / Dashboard / Readiness, so engineering inherits a plan, not prose.

2. Slack gateway — knowledge loop for non-CLI stakeholders​

The differentiated surface. Stakeholders who won't install a CLI can DM or @-mention a bot in the Slack workspace they already use; one host runs the bridge against their machine and pays the LLM bill.

The loop: retrieval over your PKB → automatic mra-ask when retrieval isn't enough → escalation to a human IT contact → absorb the answer into a KnowledgeAtom → approval gate → future questions hit the cached answer.

This loop is what makes PMK more than a PRD generator or a wiki bot.

3. Desktop app — early shell (secondary)​

Electron GUI over the same CLI surface (chat panel + worktree manager). Still early; treat as a secondary surface until parity with the Desktop PRD is reached. The CLI and gateway carry the core experience today.

Base value vs. mra-enhanced value​

PMK does two distinct things depending on whether multi-repo-agent (mra) is installed. Both are real; just be clear which one you're getting.

FeatureWithout mraWith mra
pmk propose / ask / caseRAG over your docs/ onlySame + grounded in real module / endpoint names from your repos
traceability.js / Confluence syncFull functionalityFull functionality
Slack gateway PKB retrievalWorks on your existing markdownSame, plus auto-loaded mra:--all summaries on first turn
Gateway mra-ask escalationNo-opBot can answer code-level questions before tagging a human
Knowledge atom absorb / retrievalWorksWorks

In short: base value = traceable PM docs + Slack Q&A grounded in markdown. mra-enhanced value = the above + code-aware answers. Both surface in the same CLI / gateway; the difference is what the model can ground in.

30-day adoption path​

The fastest way to feel the value is to follow this order. The earliest steps don't need mra; mra-enhanced value comes online in Week 2.

  • Day 1 — pmk propose produces a real PRD in docs/prds/. Same-day output, no infra needed.
  • Day 2–3 — Author a second PRD. traceability.js now has links to validate; the dependency graph and orphan report start earning their keep.
  • Week 1 — Stand up the Slack gateway against one channel. Today's flow lives in the gateway lifecycle deep-dive, with a 30-minute walkthrough in the onboarding guide; pmk gateway doctor + gateway start --dry-run make the first install verifiable before live traffic. Seed your existing PKB.
  • Week 2 — Run the full knowledge loop end-to-end: ask → mra-ask → escalation → absorb → reuse. By this point, the second PM who asks a previously-escalated question gets the cached answer without anyone re-typing.

What you don't get​

  • A project management tool. Tickets still live in Linear / Jira / whatever.
  • A wiki replacement. This kit works with Confluence; it doesn't try to replace it.
  • Auto-generated architecture. Humans still decide, ADRs still get written, migrations still get planned — the kit just gives those acts a shared shape.
  • A hosted SaaS. The CLI and gateway run on your machine against your repos. Nothing leaves the host.
  • High-precision vector knowledge search. Retrieval is intentionally BM25 over markdown — local, cheap, explainable. If you need semantic search at corpus scale, this isn't that product.

Who it's for​

  • Engineer-PMs / SAs / staff engineers already living in Git, ADRs, and migration plans.
  • Teams planning or executing a platform migration — traceability + Strangler Fig + handoff templates pay off fastest here.
  • Non-engineer stakeholders — not as CLI users, but as participants in the Slack gateway loop.
  • Small or internal-tooling teams that accept a host-run, local-first, non-SaaS deployment model.

Next​