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 externalizes agent state into plain markdown files that survive session boundaries, model swaps, and process restarts. Because the system is built on files rather than a framework, it works with any agent runtime that can read and write files — a thin adapter is all that is needed to tell a specific runtime which file to load first and what to write back when work is done.

​
Supported runtimes

Each adapter specifies the entry file the runtime auto-loads and the path to the adapter definition inside the Continuity skill:
RuntimeEntry fileAdapter
Claude CodeCLAUDE.mdadapters/claude.md
OpenAI CodexAGENTS.mdadapters/codex.md
HermesAGENTS.mdadapters/hermes.md
OpenClawAGENTS.mdadapters/openclaw.md
Agent ZeroAGENTS.mdadapters/agent-zero.md
The entry file is the single file your runtime loads first. It declares the state topology — which files exist, in what load order, and what governance rules apply. Everything else follows from there.

​
Generic runtime setup

If your runtime does not appear in the table above, the four-step pattern works for any agent that can read markdown files:
1

Copy templates to your workspace

Copy the files you need from the Continuity repo into your agent workspace. Start with the Minimum tier and add files as your agent’s loop requires them.
cp continuity/{AGENTS,IDENTITY,SELF,SOUL,STATE,INIT,TODO,SNAPSHOT}.md /your/workspace/
2

Read AGENTS.md, then INIT.md

Point your agent at AGENTS.md as the entry file. On first boot, run the INIT.md nine-stage reconstitution procedure. It inspects the workspace, creates any missing files, and declares READY when the state chain is valid.
3

Load the state chain

Before each action, load the smallest sufficient chain from root to the active task scope. Do not load every file by default — load only what the current task needs.
Global state → project state → subsystem state → task state → current action
4

Write back changed state

After every meaningful action, update the files whose claims are no longer accurate. At minimum: update STATE.md with new reality, move completed items from TODO.md to completed/, and refresh SNAPSHOT.md. No stale state — if reality changed, state changes.

​
Auto-injection

Some frameworks inject files into the agent’s context automatically on every session. AGENTS.promptinclude.md in the Continuity repo is the canonical auto-injection config. It lists files using @filename syntax. Any file listed there is pulled into the system prompt without the operator needing to reference it manually each session. 
# AGENTS.promptinclude.md — excerpt

# Required tier (auto-loaded)
@AGENTS.md
@IDENTITY.md
@SOUL.md
@SELF.md
@USER.md
@INIT.md
@MEMORY.md
@TOOLS.md
@STATE.md
@CONTEXT.md
@PLAN.md
@TODO.md
@SNAPSHOT.md

# Recommended tier (uncomment to enable)
# @KNOWLEDGE.md
# @INFRASTRUCTURE.md
# @POTENTIAL.md
# @CONTEXT_BUDGET.md
# @MISSIONS.md
# @TASKS.md
# @LEARNINGS.md
# @WARNING.md
# @HEARTBEAT.md
# @EVENTS.md
# @BOOTSTRAP.md
# @BOOT.md
Auto-injection is opt-in and context-budget-aware. Only include files your agent actually needs on every turn. Uncommenting every entry in AGENTS.promptinclude.md on a large workspace will exhaust the context window before the agent can act.
Runtimes that support auto-injection of this kind include:
  • Cursor — reads .cursorrules automatically on every turn
  • Claude Code — reads CLAUDE.md from the project root (and nested CLAUDE.md files in subdirectories)
  • Hermes — auto-injects SOUL.md, MEMORY.md, and USER.md from the Hermes home path (~/.hermes/)
  • Agent Zero — uses AGENTS.promptinclude.md to inject files into the system prompt automatically
For runtimes without native auto-injection via AGENTS.promptinclude.md (Codex, OpenClaw, CLI-based agents), reference the state files explicitly in your entry file or wrapper script. Claude Code uses its own CLAUDE.md auto-load mechanism rather than AGENTS.promptinclude.md.

​
Adapter pages

Claude Code

Wire Continuity through CLAUDE.md. Auto-loads from project root; supports hooks for pre/post-action validation.

OpenAI Codex

Wire Continuity through AGENTS.md. Codex composes instructions across global, repository, and directory scopes.

Hermes, OpenClaw & Agent Zero

Adapter notes for Hermes (progressive discovery), OpenClaw (SKILL.md security gates), and Agent Zero (prompt-include system).

Core Tiers

Not sure which files to copy? Start here — pick the smallest tier that fits your agent’s loop.

Build docs developers (and LLMs) love

Get started for freeTalk to us