decision/retry-backoff-strategy.md0.96
session/2026-03-29-payment-service.md0.91
research/idempotency-patterns.md0.84
For agents
Context that's always ready to hand
Your agent is powerful — but every session, it starts from zero. Temper gives agents what they lack: a persistent, structured, searchable context layer that makes the throughline legible.
The problem
Powerful but forgetful
Claude Code, Cursor, Windsurf, Copilot — these tools can write code, analyze architecture, and plan implementations. But they carry nothing between sessions. No memory of yesterday's decision. No awareness of the constraint that shaped today's approach. No sense of where the project is headed or what's already been tried.
You compensate by writing CLAUDE.md files, pasting context at the start of sessions, and manually steering the agent through decisions it should already know about. This works — until the project grows, the decisions multiply, and the context you need to inject exceeds what you can hold in your head.
The agent needs the same throughline that you carry unconsciously: what we're building, why, what we've decided, and what's deferred. That throughline needs to be structured, searchable, and persistent. It needs to be a vault.
Three pathways
One vault, three interfaces
Agents reach the vault through whichever interface suits their integration model. The same structured knowledge, accessed three different ways.
CLI
temper warmup, temper search, temper session save — agents that can run shell commands get full vault access. Claude Code hooks call temper warmup automatically at session start.
MCP Server
The Model Context Protocol server exposes vault operations as structured tools. Agents query temper.search, read resources, and write session notes through the protocol — no shell access required.
Skill File
temper skill install generates a Claude Code skill that teaches the agent your vault's structure, available commands, and workflow conventions. The agent learns how to use temper as a first-class capability.
Session pre-warming
Start every session informed
Add a startup hook to your project, and every new agent session begins with context — not a blank slate. The agent reads what happened, what's in progress, and what decisions shape the current work before writing a single line of code.
// .claude/settings.local.json
{
"hooks": {
"SessionStart": [{
"matcher": "startup",
"hooks": [{
"type": "command",
"command": "temper warmup --context myapp"
}]
}]
}
}
Every startup injects:
In-progress tasks with mode and effort
Last 5 session summaries
Full content of the most recent session
MCP server
Direct agent integration
The MCP server exposes the vault as a set of structured tools that any MCP-compatible agent can call. No file system access needed — the agent queries the vault through the protocol and gets structured results back.
Skill generation
Teach the agent your workflow
The generated skill file is a Claude Code skill that describes your vault's structure, your available commands, and your workflow conventions. The agent doesn't need to be told how to use temper — the skill makes it a first-class capability, the same way a plugin teaches an editor a new language.
$ temper skill install
Generating skill from vault structure...
→ 3 contexts, 8 goals, 47 tasks
→ Modes: build / plan · Effort: small / medium / large
→ Session template: goal / happened / decisions / next
Installed to ~/.claude/commands/temper.md
The skill evolves with your vault. As you add projects, refine templates, and develop conventions, temper skill install regenerates to match. The agent always has current instructions.
The philosophy
If it can read files, it can use temper
Temper doesn't require a specific agent, a specific IDE, or a specific workflow. The vault is markdown. The search is semantic. The protocol is MCP. Any tool that can read files — or call an MCP server — gets the full throughline.
This is deliberate. The knowledge base is the unit of value, not the tool. Your vault works with Claude Code today and whatever comes next tomorrow. No vendor lock-in, no proprietary format, no walled garden. Your context belongs to you.
Agents work best when humans temper the context. Temper's session-over-session workflow gives builders and agents the same throughline — what we're building, why, what's decided, and what comes next.