Built at BetaCraft — PM Operations

Project management,
operated by AI.

23 skills that handle the full PM lifecycle — SOW to dashboard, transcript to decision log. You oversee. Claude operates.

23
Total skills
4
Things you do
~20
Min weekly overhead
betacraft.github.io/pm-project-alpha
Project Alpha dashboard — generated by pm-* skills
DEC-004 logged
Vendor decision · sourced from transcript
How it works

You decide.
Claude operates.

The system handles everything that doesn't require your judgment. You handle everything that does.

01
Run install once
One command creates your project folder, initialises Git, creates a GitHub repo, and connects the Obsidian knowledge graph. Takes 2 minutes.
02
Answer build-sow questions
Claude interviews you conversationally. 10 questions. You answer naturally. It writes the full SOW, locks it as Tier 1 Canonical, extracts requirements.
03
Paste transcripts after calls
Paste any Fathom transcript. Claude extracts every change, decision, and action item. You review the HITL document. Say approved.
04
Start and end sessions
Say "start session" — Claude reads everything and briefs you. Say "end session" — it updates the index, logs reasoning, commits, pushes.
Project creation
$bash install.sh
Project name:project-alpha
Client:the client
Org:1) BetaCraft
Created
Project folder✓ project-alpha/
Git repo✓ initialised
GitHub✓ betacraft/pm-project-alpha
Obsidian MCP✓ connected
23 skills✓ installed
System architecture

Every document
knows its place.

Three layers. Clear rules. Nothing is ever lost, overwritten, or misattributed.

Tier 1
Canonical
Frozen. Never edited. Everything else inherits from these.
Statement of Work
Signed contract
Baseline requirements
Canonical vocabulary
If anything conflicts with Tier 1 — the conflict is the error, not the SOW.
Tier 2
Operational
Inherits from Tier 1. Changes require HITL approval.
Requirements (REQ-*)
Deliverables (DEL-*)
Decision log
Architecture docs
Hotsheet · Index · Strategy
Conflicts with Tier 1 go to CONFLICTS.md — never silently resolved.
Tier 3
Archive
Lineage only. Never cited in outputs. Never deleted.
Meeting transcripts
Email threads
Change events
Research materials
Dashboard versions
The audit trail. Everything is traceable. Nothing is overwritten.
Project folder — at a glance
Governing layer
Tier 1 + 2
shared/
· PROJECT_INDEX.md Master map · read first every session
· DECISIONS.md Sequential DEC numbers · append-only
· HOTSHEET.md Blocking · watching · resolved
· VOCABULARY.md Tier 1 · locked with SOW
· CONFLICTS.md Never silently resolved
reference/
· sow.md Tier 1 · frozen at signing
· contract.md Tier 1 · frozen at signing
Working layer
Tier 2
requirements/
· _baseline.md Tier 1 · original SOW requirements
· REQ-001-[title].md Current requirements · versioned
deliverables/
· DEL-001-[title].md Delivery specs · verified against code
architecture/
technical/ · strategic/
workspaces/aakash/
· session-log.md What + why · append-only
· proposed-decisions.md Staging before shared/DECISIONS.md
Archive layer
Tier 3
comms/
emails/ · slack/ · meetings/
attachments/ Date-stamped external docs
· summary.md · todos.md
change-log/
· YYYY-MM-DD-[event].md Every change · dated · sourced
research/
competitive/ · technical/ · client/
dashboard-versions/
· YYYY-MM-DD-v[N].html Every published dashboard · permanent
T1
Tier 1 governs everything
If a Tier 2 document conflicts with the SOW, Tier 2 is the error. Resolved in CONFLICTS.md, never silently fixed.
Nothing changes without a trace
Every decision has a DEC number. Every change has a source quote. Every session has a reasoning log.
Knowledge graph underneath
Every document links to related documents via wikilinks. Obsidian MCP indexes all of it — Claude navigates by relationship, not keyword search.
Setup

One command.
Everything ready.

No configuration files. No manual setup. No connecting tools one by one. Run one script — your entire PM infrastructure is live in under 3 minutes.

Project folder + Git repo
Self-contained. One folder per project. No parent workspace to manage.
GitHub repo + Pages branch
Client dashboard gets a permanent URL from day one.
Obsidian knowledge graph
MCP server configured in Claude Desktop. Active immediately after one restart.
23 skills installed
Every skill loaded and ready. No copy-paste, no manual file creation.
Organisation branding
Choose your org at setup — dashboard inherits logo and colour automatically.
bash install.sh
$ bash install.sh
BetaCraft PM System — New Project Setup
─────────────────────────────────────
Project name: project-alpha
Client name: Enterprise Client
Organisation: 1) BetaCraft
Project folder created
Git initialised + first commit
GitHub repo: org/pm-project-alpha
GitHub Pages branch created
obsidian-mcp-server installed
Claude Desktop config updated
23 skills ready
org.json — BetaCraft branding set
▶ Restart Claude Desktop to activate
▶ Open project-alpha/ in Cowork
▶ Say: "build sow"
Ready in 2 min 41 sec
Existing projects?
Run bash migrate.sh ~/path/to/project — adds the full system to any existing project without touching what's already there. Full backup created first.
Intelligence layer

The more it grows,
the smarter it gets.

At month 6 with 80+ documents, most PM systems get slower and less accurate. This one gets faster and more precise.

01
Every document links to related documents
Requirements link to their SOW clause, change events, decisions, transcripts, and deliverables. Obsidian indexes every link automatically.
02
Claude navigates by relationship, not keyword
Instead of reading 80 files linearly, Claude follows the graph. "Full history of REQ-003" traverses 6 linked documents — precise, in seconds.
03
Obsidian runs silently in the background
You never open it. Never see the graph. It is pure infrastructure — configured automatically by install.sh. One restart. Always on.
04
Every session runs drift detection
The graph knows which documents inherit from the SOW. session-start checks every Tier 2 document against Tier 1 — automatically, every morning.
Knowledge graph — REQ-003 traversal
REQ-003
Core requirement
sow.md
Tier 1
DEC-004
Decision
DEC-007
Decision
CHG-2025-05
Change event
Transcript
May 21
DEL-003
Deliverable
Natural language query
"Full history of REQ-003 — every decision, change, and transcript"
6 documents traversed · Complete history in 1.2s
80+
Documents at month 6
Without graph: slow, context-heavy, connections missed. With graph: fast, precise, always accurate.
100%
Abstracted from user
You never open Obsidian. Never see the graph. It runs silently — configured once by install.sh.
Every
Session starts with drift check
The graph knows document relationships. Conflicts with Tier 1 surface automatically — before you start work.
Local
Runs entirely on your machine
No cloud, no external server, no subscription. The MCP server reads your project files directly.
The real impact

PM admin time,
reduced by design.

Not "AI will help you work faster." Specific tasks. Specific times. Before and after.

Before pm-*
4–6 hrs / week
Writing meeting notes after calls
45–60 min
Updating requirements from transcript
30–45 min
Preparing pre-call status brief
20–30 min
Generating client status report
45–90 min
Tracking decisions across emails
30 min+
Hunting context from old threads
Unpredictable
Publishing a client dashboard
1–2 hrs
After pm-*
~17 min / week
Paste transcript, say "approved"
5–10 min
Claude extracts + updates automatically
0 min
Say "call brief"
60 sec
Say "client report"
60 sec
Logged automatically from transcript
0 min
Say "query: [anything]" via knowledge graph
< 5 sec
Say "publish dashboard"
30 sec
As projects grow past 50 documents — the knowledge graph keeps retrieval instant. Linear reading degrades. Graph traversal does not.
Skill library

23 skills.
One system.

Philosophy

Built on three
non-negotiable rules.

01
Nothing changes without a trace
Every decision gets a DEC number. Every change gets a source quote. Every session gets a reasoning log. Months later you can answer "why does this say that?"
02
Tier 1 governs everything
The SOW is Tier 1. If anything conflicts with it — the conflict is the error, not the SOW. Drift is detected every session start, automatically.
03
No single memory as bottleneck
The session routine — start, end — keeps the project state current so any PM can pick up exactly where things left off with zero briefing.
Weekly workflow

Your week,
redesigned.

What used to take hours of admin now takes minutes — or happens automatically.

Monday morning 30 seconds

Say "start session". Claude reads the governing layer, runs drift detection, checks vocabulary, and briefs you on everything that happened since Friday.

session-start drift detection vocabulary check
Before a client call 60 seconds

Say "call brief". Claude reads all project context and generates a structured brief — status, changes since last call, open items, risks, talking points.

call-brief query-knowledge
After a client call 5-10 minutes

Paste the transcript. Claude extracts every change, decision, and action item. You review the HITL document and say "approved". Claude does the rest.

process-transcript log-decision update-hotsheet
Friday end of day 30 seconds

Say "end session". Claude updates the index, appends the session log with reasoning, checks for uncommitted changes, and pushes everything to GitHub.

session-end update-index
Your total weekly overhead
Start/end session (×5 days)
5 min
Call briefs (×2 calls avg)
2 min
HITL review after transcripts
10 min
Total
~17 min
Everything else
Claude does it.
+ Requirements tracking
+ Decision logging
+ Dashboard generation
+ Comms sync from Gmail + Slack
+ Drift detection
+ Git commits and pushes
Built for real projects

Used across active
client engagements.

"Taking a step back and sharpening the axe like this will make the future much better and easier to tackle. I truly love it."
Client · Enterprise SaaS
Client · Enterprise Platform Care Intelligence Platform
"The entire project — SOW to meeting transcripts to codebase — everything now lives together. What was fragmented before is now end-to-end."
Engineering Lead · Agency
Lead Engineer · the agency
Active projects
Enterprise Platform v2 Active
Project Alpha Active
Social App Build
Telecom Portal Discovery
Retail Intelligence Active
Active client roster · June 2026