Claude Code Project Manager Playbook
Turn Claude Code into the project manager that runs your Kanban across every project you touch. Two markdown files per project, one optional plugin, strict WIP caps, ICE-scored prioritisation, and a Recurring-tasks layer that auto-surfaces overdue work at session start. Five English phrases run the whole system. Set up in 30 minutes.
Get the full Claude Code Project Manager Playbook playbook
Enter your email. We'll send you the link instantly. Point your Claude Code at it and run the full audit in 20 to 40 minutes.
- What it does: installs a persistent project-management system into any Claude Code project. Two markdown files (`CLAUDE.md` and `actions.md`) plus an optional plugin (`ec-kanban-core`) with a SessionStart hook.
- Output: a working Kanban board with P0/P1/P2 caps, ICE scoring with 80/20-weighted effort, task-owner tags ([Claude], [you], [Hybrid]), a Recurring-tasks register that surfaces due items at session start, and five natural-language triggers mapping to five discrete behaviours.
- Runs in: about 30 minutes, end-to-end, from paste to installed and running on one project. Multi-project install pattern takes another 10 minutes per additional folder.
- You need: Claude Code installed, Python 3.9 or newer, git in the project folder. No paid tools. No external API keys.
- For: solo founders running multiple workstreams, agency operators running multi-client work, engineering leads and solo developers juggling side projects, anyone who opens Claude Code daily and wants continuity across sessions.
- Cost: free. No external APIs. No SaaS. The only spend is whatever you already pay for Claude Code (Pro, Max, or API).
- Built by: Nishant Kapoor at EntireCommerce AI. Tested across eight live projects before packaging.
What this playbook does
Open Claude Code on any given day, describe your project in a paragraph, do some work, close the session. Open a new session the next day and describe the project again. Context evaporates between sessions. The to-do list lives in a Notion doc that gets updated once a week. The weekly review is a recurring calendar block that gets skipped half the time. Claude becomes a smart executor and a forgetful one. Project management stays in the operator's head.
This playbook replaces that with a persistent system. `CLAUDE.md` holds the project brief Claude reads on every session start: what the project is, who it's for, wired-up tools, operating rules, brand voice. `actions.md` holds the live work state: a Recurring section at the top, a Kanban board in the middle with strict caps (P0 ≤ 5, P1 ≤ 10, P2 ≤ 15), and a Completed log at the bottom. Every board item carries an ICE score and a task-owner tag. An optional Claude Code plugin (`ec-kanban-core`) ships a SessionStart hook that reads every `actions.md` in the project tree and surfaces overdue recurring work at the top of the session. Five English phrases run the system: "what's next?" triggers the triage pass, "let's wrap up" triggers the reprioritisation pass, and three more handle the Recurring layer. Claude becomes the operator of the board, and the board persists across every session because it lives in two markdown files under git version control.
What you get
A populated project folder with Claude Code wired to run the Kanban on every session. Eight concrete deliverables:
- Persistent project brief in CLAUDE.md. Loads on every session start. Holds project context, operating rules, wired-up tools, brand voice. Every future session starts with full context loaded from the first message.
- Seeded Kanban board in actions.md. Your current work organised into P0 / P1 / P2 tiers, each item ICE-scored and tagged. Strict WIP caps enforced (5 / 10 / 15). Blockers auto-promoted.
- Recurring-tasks register. Weekly reviews, monthly audits, quarterly refreshes tracked with cadence and last_run metadata. Section sits above the Kanban and doesn't count against caps.
- Optional ec-kanban-core plugin. SessionStart hook wired into ~/.claude/settings.json. Surfaces overdue recurring items at the top of every future session. Python, 200 lines, zero dependencies.
- Natural-language trigger vocabulary. Five English phrases mapped to five behaviours. Documented in CLAUDE.md so Claude responds consistently across sessions. No slash commands to memorise.
- ICE scoring rubric with weighted effort. 80 percent your effort plus 20 percent Claude's effort. Biases the board toward work that draws minimally on your time.
- Task-owner tag system. [Claude] for work Claude can run end-to-end. [You] for work that requires your hands. [Hybrid] for review-and-approve loops. The board stays honest about who is actually doing the work.
- Git-backed audit trail. Every session commits CLAUDE.md and actions.md changes. The Completed log plus git history gives you a year-end retrospective artefact for free.
How the playbook flexes
Pick the install mode that fits where you are today. Claude confirms your choice before writing anything.
| Mode | When it applies | What changes |
|---|---|---|
| Lightweight | One solo project, no recurring work yet | Skip the plugin install. CLAUDE.md plus actions.md only. Kanban on top. Manual Recurring check via "what's due?". |
| Standard | Multi-project operator, recurring work exists (weekly reviews, monthly audits) | Install the plugin. SessionStart hook fires automatically. Recurring items surface at session start and compete for P0 slots on merit. Default install. |
| Heavy | Many active projects in parallel (multi-client agency, portfolio operator) | Standard mode per project. Plus layered plugins on top of ec-kanban-core for client-specific workflows (ec-gtm-toolkit for marketing engagements, per-client ops plugins for heavy integrations). |
| Dry-run | You want to see the plan before writing any files | Claude prints the full install plan: directory structure, files, plugin wiring. Zero files written until you approve. |
Tools the system uses
Required.
- Claude Code as the execution layer
- git for version control of CLAUDE.md and actions.md
- Python 3.9 or newer for the optional SessionStart hook
Optional.
- GitHub or any git host for remote backup of the markdown files
- pandoc if you want to export weekly reviews as Word docs
- Any existing Claude Code skills you already use (the system plays nicely with /weekly-review, /gtm-audit, /cro-review, /claude-md-improver, and every marketing skill in the EC library)
Explicitly not needed. No Asana. No Linear. No Notion. No Trello. No Jira. No paid APIs. No SaaS. The whole system runs on Claude Code plus two markdown files plus one Python script.
The Recurring-tasks layer: what makes this different
Most project-management systems treat recurring work as a side feature. You set up a weekly review as a calendar event. You set up a monthly audit as a recurring task in Asana. Neither system knows how many cycles have slipped or how the recurring work should compete for your attention against the ad-hoc work. The result: recurring work silently rots (six weeks since the last CLAUDE.md audit) or the ad-hoc work silently rots (two weeks since anyone shipped a client deliverable because the founder was busy running weekly reviews).
This system fixes both failure modes. Recurring items live in a dedicated register with `cadence` and `last_run` metadata. A SessionStart hook parses the register at the top of every session and surfaces any item where `last_run + cadence` is less than or equal to today. When surfaced, a recurring item gets ICE-scored against the current P0 queue. If it genuinely outranks an in-flight client deliverable, it earns a P0 slot (and the weaker item drops to P1). If it doesn't, the recurring work is overdue but the bottleneck this cycle lives elsewhere. You decide explicitly: pull it in, or skip this cycle (which advances last_run by one cadence). Nothing slips silently. Nothing gets prioritised by calendar alone.
The mechanism is boring. The effect is durable. Across eight live projects tested for a month, the Recurring register caught six cadence slips that would otherwise have gone unnoticed for weeks.
Ready to install Claude Code as your project manager?
Enter your email. We'll send you the bundle link. Paste it into Claude Code and you'll have the full system installed in about 30 minutes.
Questions we get asked
What is the Claude Code Project Manager Playbook?
It is a persistent project-management system installed inside any Claude Code project. Two markdown files per project (`CLAUDE.md` for context, `actions.md` for work state), plus an optional Claude Code plugin that ships a SessionStart hook. The Kanban board in `actions.md` has strict WIP caps (P0 ≤ 5, P1 ≤ 10, P2 ≤ 15). Every item is ICE-scored with a weighted-effort formula (80 percent your time plus 20 percent Claude's time). Five English phrases run the whole system. Claude becomes the operator of the board across every session.
Is this free?
Yes. The playbook itself is free. No external tools. No paid SaaS. The only spend is whatever you already pay for Claude Code (Pro, Max, or API). No account required with us, no sales call.
How is this different from Asana, Linear, or Notion?
Those tools assume you open their UI, review the backlog, and reprioritise manually. Claude Code Project Manager flips the flow: the backlog lives in two markdown files Claude reads on every session start, so prioritisation happens every time you touch the project. You say "what's next?" and Claude triages in one pass. You say "let's wrap up" and Claude reprioritises, logs completions, and pushes to git. Project management stops being a separate app and becomes the default behaviour of your Claude Code session.
How long does the install take?
About 30 minutes for the first project from paste to installed and running. 10 to 15 minutes per additional project once you know the flow. Most of the time is the interview where Claude asks about the project, your recurring work, and any operating rules. File creation takes under 5 minutes.
Do I need to be technical?
You need Claude Code installed and a project folder under git version control. That is the entire technical bar. You do not need to write code, configure servers, or manage databases. The optional SessionStart hook is Python 3.9 or newer, which is already installed on macOS and most Linux distros. The install prompt handles the wiring.
What is the ec-kanban-core plugin?
ec-kanban-core is a small Claude Code plugin that ships alongside this playbook. It installs a SessionStart hook that parses the Recurring section of every `actions.md` in your project tree at session open and surfaces any overdue items at the top of the session. Hook is about 200 lines of Python. Zero external dependencies. Install is optional: you can run the whole system manually without it if you prefer to check the Recurring register yourself.
What are the five English phrases?
"What's next?" triggers the triage pass: Claude parses the Recurring register, ICE-scores due items against the current P0 queue, slots them at earned rank, and returns the top 5 items to work on. "What's due?" gives a read-only view of the Recurring register. "Let's do the weekly review" (or any specific recurring workstream) jumps straight to execution and updates last_run when done. "Skip the weekly review this cycle" advances last_run by one cadence. "Let's wrap up" reprioritises, logs completions, and pushes to git at end of session.
Does this work across multiple projects?
Yes. The system is designed for operators running multiple projects in parallel. Install the playbook once per project folder. The SessionStart hook reads every `actions.md` in the project tree, so one hook installation covers every project. The five English phrases work identically across every project. The multi-project install pattern is the Standard or Heavy mode, picked during the install interview.
Who built this playbook?
Nishant Kapoor, founder of EntireCommerce AI. Chartered Accountant (EY), INSEAD MBA, ten-year DTC operator. Scale stories include 185% revenue growth on a UK luxury resale brand in two years, a consumer wearable $0 to $3.5M in eighteen months, and consumer electronics $0 to $5M in two years. This playbook was tested across eight live projects before packaging.
Can I run this myself, or do I need you?
You can run it yourself. The install prompt handles the mechanical work. Your job is to answer a few context questions about the project and review the final output. If you want a senior operator to run this plus the ongoing weekly cadence for you on a live engagement, book a call.