Give your AI coding agents persistent memory and shared team standards.
AI coding assistants forget everything between sessions. They re-read your code every time, miss your team's conventions, and repeat the same mistakes. On a team, it gets worse — every developer's agent behaves differently.
agentskel fixes this. It gives every agent on your team a persistent memory of your codebase, a set of rules they always follow, and structured workflows for common tasks like building features, fixing bugs, and reviewing code. Install it once — no changes to your application code.
| Without agentskel | With agentskel |
|---|---|
| Agent re-reads 40 files every session | Agent remembers your code between sessions |
| Agent doesn't know your team's conventions | Agent learns and enforces your team's standards |
| Each developer's agent behaves differently | Everyone's agent follows the same rules |
| Past mistakes get repeated | Agent remembers what went wrong and avoids it |
| Cross-platform apps drift out of sync | Shared business logic stays consistent across platforms |
Your team's shared knowledge lives in a Git branch inside your repo. Plain markdown files, versioned alongside your code, shared via git push. No database, no API key, no vendor lock-in — just Git.
# Claude Code
/plugin install agentskel
# Gemini CLI
gemini extensions install https://github.com/ahmadulhoq/agentskel
# Cursor / GitHub Copilot / Windsurf / Codex — see INSTALL.mdOpen your project in your AI tool and say one of the following:
Single project?
"Set up agentskel on this project."
A folder with multiple projects inside (e.g. a backend folder + a mobile app folder)?
"Set up agentskel as a workspace."
Multiple projects that share the same business logic and need to stay in sync?
"Create a blueprint here."
Not sure which fits? See docs/INSTALL-MODES.md.
After setup, connect your team's Jira, Confluence, and GitHub so the agent participates in your work processes:
Capture the team roster:
"Set up the team."
Enable ticket-based workflows (read tickets, update status, reassign to QA):
"Set up Jira."
Enable publishing ADRs, postmortems, and specs to Confluence:
"Set up Confluence."
Each setup is independent — use any subset. Requires the Atlassian MCP connected in your AI tool. See docs/ATLASSIAN-SETUP.md for connection details and docs/TEAM-COORDINATION.md for how the three layers (memory, Jira, Confluence) work together.
If your teammate already set up agentskel, clone the project and run:
./scripts/install-agent.shYour agent now has access to everything the team has learned.
A shared memory of your codebase. The agent reads every file once, indexes every class and function, and remembers how modules connect. Next session, it doesn't need to re-read.
Rules every agent follows. Security guardrails, coding conventions, planning before coding, verifying before shipping. The same rules apply whether you use Claude Code, Cursor, Copilot, or Gemini.
Workflows for common tasks. Plain-English commands like:
- "Develop a feature for X" → plans, branches, writes tests, implements, opens PR
- "Fix this bug" → reproduces, writes a failing test, finds root cause, fixes
- "Map this codebase" → builds a full index the agent uses forever after
- "Check dependencies" → audits for outdated or risky packages
Cross-platform coordination (optional). If your team has a backend and a mobile app that share business logic, agentskel keeps them in sync automatically. When the backend changes an API contract, the mobile agent sees it at next session start.
Claude Code · Cursor · GitHub Copilot · Windsurf · Codex CLI · Gemini / Antigravity
One setup, every tool. Your teammates don't all have to use the same AI.
A system prompt is per-developer, per-tool, and resets every session.
agentskel's knowledge lives in Git — shared across the team, works across every tool, and persists indefinitely. When your tech lead documents an architectural decision, every agent on every tool knows about it immediately. When one agent learns from a production bug, no agent ever makes the same mistake again. A system prompt stays flat. agentskel compounds.
agentskel gets better over time. Your agent detects when a new version is available:
"Sync this project with the latest skeleton."
The agent walks through the changes, picks the ones that apply, and opens a PR.
Install to the canonical machine-level location:
git clone https://github.com/ahmadulhoq/agentskel ~/.agentskel/skeletonThat's it. setup-skeleton and sync-skeleton find it automatically from that path —
no configuration needed. The skeleton auto-updates (git pull) each time a
workflow runs.
If you already have agentskel cloned elsewhere, it still works — set
Skeleton Path in .memory/CONFIG.md to your existing clone path and
the auto-resolution will use it first.
v1.66.0 — see CHANGELOG.md for what's new.
See CONTRIBUTING.md. Skills and workflows are plain Markdown — no code required.
MIT — see LICENSE.