Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/bitwikiorg/continuity/llms.txt

Use this file to discover all available pages before exploring further.

Continuity’s tiers are suggested starting templates, not requirements. They exist because different agents need different amounts of state — a simple task-runner needs almost nothing; a recursive multi-agent orchestrator needs the full inventory. Pick the smallest tier that gives your agent the continuity it actually requires, then add files only when your loop genuinely needs them.

​
The four tiers

TierFile countFor
Minimum8 filesSimple agents with a single focused loop
Standard16 filesMost agents — adds memory, tools, planning
Full27 filesOperational agents with runtime discipline
Advanced30 filesRecursive multi-agent systems

​
Minimum

The smallest configuration that produces a functional agent with persistent identity and task tracking. An agent booted on Minimum files can declare READY, remember who it is, and maintain a task list across sessions. Files: AGENTS, IDENTITY, SELF, SOUL, STATE, INIT, TODO, SNAPSHOT 
cp continuity/{AGENTS,IDENTITY,SELF,SOUL,STATE,INIT,TODO,SNAPSHOT}.md /your/workspace/
What this gives you:
  • AGENTS.md — bootloader, runtime parameters, project guard
  • IDENTITY.md + SOUL.md + SELF.md — full identity stack (who, why, how)
  • STATE.md — current operational truth
  • INIT.md — 9-stage reconstitution procedure
  • TODO.md — task index
  • SNAPSHOT.md — restoration anchor
What this omits: No user model, no memory, no tools inventory, no planning layer, no safety gates, no heartbeat. For simple agents where none of those are needed, Minimum is the right choice.

​
Standard

The practical default for most agents. Adds the user model, memory, tools awareness, planning, and a learning layer on top of Minimum. Files: everything in Minimum, plus USER, MEMORY, TOOLS, PLAN, CONTEXT, TASKS, LEARNINGS, HEARTBEAT 
cp continuity/{AGENTS,IDENTITY,SELF,SOUL,STATE,INIT,TODO,SNAPSHOT}.md /your/workspace/
cp continuity/{USER,MEMORY,TOOLS,PLAN,CONTEXT,TASKS,LEARNINGS,HEARTBEAT}.md /your/workspace/
What the new files add:
  • USER.md — model of the human operator; the agent adapts to the person it serves
  • MEMORY.md — current memories and index to memory/ entries
  • TOOLS.md — capability registry and environment notes
  • PLAN.md — intended sequence, phases, milestones (indexes proposals/)
  • CONTEXT.md — per-turn operational snapshot (indexes journal/)
  • TASKS.md — system and housekeeping tasks separate from project work
  • LEARNINGS.md — soft rules and insights from prior sessions
  • HEARTBEAT.md — periodic check protocol
Standard is the right choice when the agent needs to remember facts across sessions, maintain a tool inventory, or follow a multi-phase plan.

​
Full

For agents operating in production environments with runtime discipline, safety constraints, and a broader capability surface. Adds infrastructure awareness, context budget management, safety gates, and a full boot/bootstrap path. Files: everything in Standard, plus INFRASTRUCTURE, POTENTIAL, CONTEXT_BUDGET, MISSIONS, BACKLOG, WARNING, BOOTSTRAP, KNOWLEDGE, EVENTS, ARTIFACTS, BOOT 
cp continuity/{AGENTS,IDENTITY,SELF,SOUL,STATE,INIT,TODO,SNAPSHOT}.md /your/workspace/
cp continuity/{USER,MEMORY,TOOLS,PLAN,CONTEXT,TASKS,LEARNINGS,HEARTBEAT}.md /your/workspace/
cp continuity/{INFRASTRUCTURE,POTENTIAL,CONTEXT_BUDGET,MISSIONS,BACKLOG,WARNING,BOOTSTRAP,KNOWLEDGE,EVENTS,ARTIFACTS,BOOT}.md /your/workspace/
What the new files add:
  • INFRASTRUCTURE.md — live substrate inventory (services, endpoints)
  • POTENTIAL.md — Structure / Potential / Void capability model
  • CONTEXT_BUDGET.md — prompt hygiene rules and context limits
  • MISSIONS.md — high-level priorities above the task layer
  • BACKLOG.md — rolling inbox for pre-triage items
  • WARNING.md — hard limits and pre-action checklist
  • BOOTSTRAP.md — first-run identity discovery (self-destructing)
  • KNOWLEDGE.md — catalog of knowledge documents (indexes knowledge/)
  • EVENTS.md — recent event summary (indexes logs/)
  • ARTIFACTS.md — generated artifact inventory (indexes artifacts/)
  • BOOT.md — lightweight CHECK → ACT → REPORT cycle for heartbeats
Full is the right choice for agents that manage infrastructure, generate artifacts, operate under safety constraints, or need to manage a large context budget across many files.

​
Advanced

For recursive multi-agent systems where one agent spawns and manages sub-agents with bounded delegated authority. Adds the approval gate, action reference, and machine-readable binding contract. Files: everything in Full, plus CHECKPOINT, AGENT_ACTIONS, BINDING.schema.json 
cp continuity/{AGENTS,IDENTITY,SELF,SOUL,STATE,INIT,TODO,SNAPSHOT}.md /your/workspace/
cp continuity/{USER,MEMORY,TOOLS,PLAN,CONTEXT,TASKS,LEARNINGS,HEARTBEAT}.md /your/workspace/
cp continuity/{INFRASTRUCTURE,POTENTIAL,CONTEXT_BUDGET,MISSIONS,BACKLOG,WARNING,BOOTSTRAP,KNOWLEDGE,EVENTS,ARTIFACTS,BOOT}.md /your/workspace/
cp continuity/{CHECKPOINT,AGENT_ACTIONS}.md continuity/BINDING.schema.json /your/workspace/
What the new files add:
  • CHECKPOINT.md — approval gate before irreversible actions
  • AGENT_ACTIONS.md — conceptual action reference for the recursive runtime
  • BINDING.schema.json — machine-readable binding contract; validates that declared mesh edges match the actual directory structure
Advanced is only needed for orchestrators that delegate work to sub-agents and need machine-verifiable authorization boundaries.

​
Guidance: when to add a file

The tier table is a starting point — the real rule is simpler:
Add a file only when your agent’s loop repeatedly reaches for state it doesn’t have.
If the agent is frequently re-discovering the same facts, add MEMORY.md. If it keeps losing track of which tools are available, add TOOLS.md. If it’s making irreversible operations without human review, add WARNING.md and CHECKPOINT.md. If it never needs to plan across multiple phases, leave PLAN.md out.
30+ files is heavy. Most agents need 3–5.The full template set exists because different loop types need different state shapes — a coding loop, a research loop, and a multi-agent orchestration loop each have different continuity needs. The templates are loop patterns, not bloat. Start with AGENTS.md + TODO.md + STATE.md and add files only when your agent’s loop actually needs them.

​
Sample configurations

The repository’s samples/ directory contains concrete, pre-configured workspaces for common agent types — a minimal coding agent, a standard research agent, and a full operational agent. Browse the domain-specific state packs for pre-wired configurations you can copy and customize.

Minimum sample

Simple coding agent — 8 files: identity, boot, and task tracking only.

Standard sample

Research agent — 16 files: adds user model, memory, tools, planning, tasks, learnings, and heartbeat.

Full sample

Operational agent — 27 files: adds infrastructure, potential, context budget, missions, backlog, safety, knowledge, events, artifacts, and boot.

Packs overview

Domain-specific packs — developer, backend, researcher, judge, ops, orchestrator, and more.
Copy a sample, fill in the {{PLACEHOLDER}} values, delete the files that don’t apply to your use case, and boot. Samples are starting points, not fixed identities.

Build docs developers (and LLMs) love

Get started for freeTalk to us