Maintains project tracking in a simple .development directory without external tools.
Install
mkdir -p .claude/skills/self-host-development-light && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/14365" && unzip -o skill.zip -d .claude/skills/self-host-development-light && rm skill.zipInstalls to .claude/skills/self-host-development-light
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Lightweight self-hosted project management using markdown files in a .development directory. Covers backlog, planning, todo, roadmap, and saved plans — everything an agent needs to orient and start working.Key capabilities
- →Provision a `.development` directory with markdown files
- →Initialize `roadmap.md` with project direction and milestones
- →Initialize `backlog.md` for unscheduled work
- →Initialize `todo.md` for active and blocked tasks
- →Initialize `adr.md` for significant decisions
- →Initialize `CAPTURE.md` as an operator inbox
How it works
This skill provisions a lightweight, self-hosted project management system using markdown files in a `.development` directory, with one file per concern like roadmap, backlog, and todo.
Inputs & outputs
When to use self-host-development-light
- →Initialize project management
- →Update roadmap
- →Track tasks in markdown
About this skill
Self-Host Development Light
Keep all project management artifacts in the repo as plain markdown. No external tools, no ticket numbers, no heavy process. This is the light form of the self-hosted development system, for a solo developer or a small team working with agents.
The unifying rule: light form = one file per concern; full
form = one directory per concern. This skill provisions the
light form. Every file below can grow independently into its
directory counterpart (see Growing a
concern); a repo at full form for every
concern is running the complete .development architecture.
Features and components are ## sections within these files —
never separate files, never IDs. The one exception to the
no-directories rule is plans/, a filing cabinet for saved
task-scoped plans.
Setup (run once)
Before doing any work, check whether setup has already been completed — if the scaffold below exists, skip to Ongoing usage.
Provision all files from day one, each as the populated stub below. No file is optional and none is deferred until needed: the stub text is part of the deliverable. It teaches the workflow to whoever — human or agent — opens the file next.
1. Create the directory structure
.development/
├── roadmap.md direction and milestones
├── backlog.md planned, not yet scheduled
├── todo.md active and blocked work
├── changelog.md shipped work
├── adr.md decisions
├── prd.md requirements
├── design.md how the system is put together
├── stewardship.md recurring maintenance
├── lessons-learned.md traps and techniques, in hindsight
├── CAPTURE.md operator inbox (operator-owned)
└── plans/ saved task-scoped plans
└── .gitkeep
2. Initialize roadmap.md
# Roadmap
Project direction: where this is going, the shape it is
converging on, and the milestones on the way. One per
project. Current priorities live here; backlog items should
trace to something in this file. Update when the destination
changes, not for every step.
## Direction
<!-- Where the project is going and why. -->
## Milestones
<!-- Coarse and ordered. Date them when they land. -->
3. Initialize backlog.md
# Backlog
Items not yet scheduled. Add new work here. Move to `todo.md`
when it becomes active.
<!-- Newest items at the top. -->
4. Initialize todo.md
# Todo
Work in progress. Keep this short — if it grows beyond a
handful of items, move deferred work back to `backlog.md`.
## Active
<!-- Items currently being worked on. -->
## Blocked
<!-- Items waiting on something. Note what they're waiting on. -->
5. Initialize changelog.md
# Changelog
Shipped work worth recording — one dated line per item,
newest first. When an item leaves `todo.md` finished, note it
here. Not every completion needs an entry; record the ones a
future reader would want to find.
6. Initialize adr.md
# Decisions
Significant decisions — architecture, tooling, direction —
one `##` section per decision, newest first: the context, the
decision, and its consequences. No numbering; sections are
the unit.
## Open questions
<!-- Decisions that need input before work can proceed. -->
<!-- Decided: "## YYYY-MM-DD — Title" sections below, newest
first. -->
7. Initialize prd.md
# Requirements
What the project must do, one `##` section per feature or
component. Keep each section to observable behavior — what a
user or caller can verify — not implementation.
8. Initialize design.md
# Design
How the system is put together: structure, key mechanisms,
and the reasoning behind them. One `##` section per area.
Task-scoped implementation plans go in `plans/`, not here;
when a plan ships, promote its durable outcome into a
section.
9. Initialize stewardship.md
# Stewardship
Standing maintenance — recurring, identical-in-kind work
(dependency updates, consistency sweeps, doc-drift checks).
One `##` section per activity: what it covers, its
guardrails, and a dated log line per pass. These sections are
never "done."
10. Initialize lessons-learned.md
# Lessons learned
Traps and techniques from doing the work — the things
that were only obvious in hindsight, and the methods
worth reusing. One `##` section per theme, concrete
enough to lift: playbooks and skills harvest this file.
11. Initialize CAPTURE.md
# Capture
Operator inbox: quick notes, ideas, and requests jotted by
the human between sessions. Operator-owned — agents read this
file and may triage items into `backlog.md` when asked, but
never author entries here.
12. Wire the agent contract
Add a pointer in the repo root AGENTS.md (create it if
absent) so any visiting agent finds the system:
## Development tracking
Project management is self-hosted in `.development/` — flat
markdown, one file per concern, no ticket IDs. Orient by
reading `todo.md`, `roadmap.md`, and `backlog.md`. Record
decisions in `adr.md` and shipped work in `changelog.md`.
13. Verify setup
Confirm every file exists and contains its stub, and that
plans/ is present with a .gitkeep. Commit the scaffold
with a message like Scaffold .development for project management.
Migrating existing documents
Setup often lands in a repo that already has
planning-flavored markdown at the root — PLANNING.md,
NOTES.md, TODO.md, ROADMAP.md, working-notes files.
Migrate them as their own commit, after the scaffold
commit: the scaffold is mechanical, the migration is
judgment, and a reviewer should see them separately.
For each candidate document:
- Classify by scope, the same rule that separates the
working files: direction →
roadmap.md, requirements →prd.md, system structure →design.md, a single task's approach →plans/<name>.md. Most working notes are task-scoped and land inplans/intact. - Move, don't copy. Use
git mvso history survives the migration, and one source of truth holds. - Extract the live work. Unchecked checklists and
"to explore" lists inside a document are backlog items
in hiding — pull each into
backlog.mdwith a pointer back to the plan. The plan stays reference material; the backlog is where work gets scheduled from. - Harvest the lessons. A trap or technique buried in
working notes ("X renders blank when exported through
Y — do Z instead") belongs in
lessons-learned.md, where the next project can find it. - Fix inbound references. Grep for the old filename
before committing —
README.mdandAGENTS.mdoften point at it.
Not everything migrates. README.md, AGENTS.md /
CLAUDE.md, and product documentation describe the
artifact, not the work — they stay where they are.
Commit the whole migration as one commit, e.g. Migrate planning docs into .development.
Ongoing usage
Orientation
At the start of each session, read three files to orient:
todo.md— what's active and what's blockedroadmap.md— direction and current prioritiesbacklog.md— what's waiting
Glance at CAPTURE.md for new operator notes. Present a
short summary to the human before starting work. This pairs
with the What's Next checklist in AGENTS.md.
Adding work
New items go into backlog.md at the top. Each item is a
short description — one or two sentences. Add context if the
item isn't self-explanatory, but keep it brief.
- **Terraform import for dave-skills repo** — Run terraform
import for the existing repo and branch protection before
first apply. See terraform/main.tf TODOs.
Starting work
Move the item from backlog.md to the Active section of
todo.md. Don't copy — move. The item should exist in
exactly one place.
Completing work
Remove the item from todo.md. If it's worth recording, add
a dated line to changelog.md. If completing it settled a
decision, record that in adr.md — the changelog says what
shipped, the decision log says why it went the way it did.
If the work taught something a future project needs — a trap
that was only obvious in hindsight, a technique worth reusing
— capture it in lessons-learned.md.
Blocking and unblocking
Move blocked items to the Blocked section of todo.md
with a note explaining what they're waiting on. When the
blocker clears, move back to Active.
Recording decisions
When a significant decision is made — architecture, tooling,
scope, direction — add a ## YYYY-MM-DD — Title section to
adr.md with the context, the decision, and its
consequences. Questions still awaiting a decision sit in the
Open questions section at the top; move them down into a
dated section once decided.
Saving plans
When a session produces a plan worth preserving — an
implementation approach, an architecture sketch, a multi-step
breakdown — save it as a file in .development/plans/ with a
descriptive name:
.development/plans/terraform-s3-backend.md
.development/plans/oidc-role-remediation.md
Plans are reference material. They don't replace the working files.
Because plans/ is the one place documents accumulate, it is
also where staleness hides. When it grows past a handful of
files, adopt the index-drift-audit skill: an
auditor-managed INDEX.md manifest plus a deterministic
reconciler that flags orphaned, missing, and stale documents
on a recurring schedule.
Roadmap, design, and plans
Three artifacts carry "how and where," at different scopes — don't let the terms collide:
roadmap.mdcarries project scope: direction and milestones. It changes when the destination changes.design.mdcarries system scope: how the thing is put together and why. It changes when the structure changes.- **
Content truncated.
When not to use it
- →When external project management tools are preferred
- →When ticket numbers or heavy process are required
- →When a full directory form for every concern is already in use
Limitations
- →The skill does not use external project management tools.
- →The skill does not use ticket numbers.
- →The skill does not support multiple files per concern in its light form.
How it compares
This skill provides a simple, markdown-based project management system directly within the repository, unlike external tools or heavy processes that require separate interfaces.
Compared to similar skills
self-host-development-light side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| self-host-development-light (this skill) | 0 | 2mo | No flags | Beginner |
| working-summary | 0 | 4mo | Review | Intermediate |
| spec-kit-workflow | 11 | 8mo | No flags | Intermediate |
| product-manager-toolkit | 32 | 7mo | Review | Beginner |
Try saying
Example prompts that trigger this skill in your AI assistant.
You might also like
working-summary
Innei
Use when the user asks for a work summary, weekly report, 工作总结,周报,working summary, or sprint/period recap. Aggregates GitHub PR/commit/issue activity from configured repos, optionally reads Linear cycle issues via MCP, honors Chinese public holidays, and produces a markdown report. Default range is
spec-kit-workflow
jmanhype
Guides specification-driven development workflow. Automatically invoked when discussing new features, specifications, technical planning, or implementation tasks. Ensures proper workflow phases (specify → clarify → plan → checklist → tasks → analyze → implement).
product-manager-toolkit
davila7
Comprehensive toolkit for product managers including RICE prioritization, customer interview analysis, PRD templates, discovery frameworks, and go-to-market strategies. Use for feature prioritization, user research synthesis, requirement documentation, and product strategy development.
f2s-req-clarify
Lands-1203
针对 PRD/需求反问直到清楚,再可用 f2s-req-tech 出技术方案;触发:需求澄清、PRD 澄清
bmad-agent-pm
alibaba
Product manager for PRD creation and requirements discovery. Use when the user asks to talk to John or requests the product manager.
product-brief
DatBeat
Create a product brief with structured process, quality checks, and system integration