Skip to main content

Documentation Index

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

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

The agent init template applies when the target will be operated, modified, or assisted by an AI agent. It does not prescribe a universal agent file tree. Instead, it determines what this specific agent target requires — which instructions exist, what tools are accessible, what context is needed — and creates only the artifacts that materially improve the agent’s ability to operate.

​
Template metadata

FieldValue
NameAgent Init Protocol
TargetAI agent, agent workspace, agent-enabled repository, automation worker, or tool-using assistant environment
PurposeInspect what an agent requires to operate in a specific target, create or configure only applicable instructions, tools, prompts, state, and validation checks, and report agent readiness
ModesActive, Dry Run, Repair, Reinitialize
CreatesAGENTS.md when agent operation applies and valid instructions are absent; SELF.md when identity, role, capabilities, or boundaries must be explicit; USER.md when operator context is required for correct interaction; TOOLS.md when tool availability, permissions, or invocation rules must be documented; STATE.md or PLAN.md when current condition or pending work must be externalized
ConfiguresAgent instructions; model or provider settings when applicable; tool definitions; prompt files; workspace directories; state or context references; expected inputs and outputs
ValidatesInstructions are target-specific and internally consistent; required tools or providers are accessible where validation is permitted; startup path, prompts, and expected outputs are clear; existing instructions are preserved or repaired without unnecessary replacement
Optional outputsAGENTS.md; SELF.md; USER.md; TOOLS.md; STATE.md; PLAN.md; SNAPSHOT.md; prompt files; tool registry; startup validation report

​
When to use

Use this template when the target is an AI agent, agent workspace, agent-enabled repository, automation worker, tool-using assistant environment, or system whose operation depends on agent instructions.

​
When NOT to use

Do not create generic agent files in every project. Agent artifacts are useful only when they materially improve operation. If the target is a software project that happens to have some agents working in it occasionally, the development project template is likely more appropriate — AGENTS.md is one conditional output of that template, not a reason to switch to the agent template.
Do not install a reusable generic instruction file without adaptation. Every artifact created by this template must be derived from the specific target. A copied, unmodified AGENTS.md from another project is not a valid initialization output.

​
Conditional artifacts

The agent template creates only artifacts the specific agent target requires. Each possible output has a condition:
ArtifactCondition
AGENTS.mdAgent operation applies and valid instructions are absent
SELF.mdIdentity, role, capabilities, or boundaries must be explicit
USER.mdOperator context is required for correct interaction
TOOLS.mdTool availability, permissions, or invocation rules must be documented
STATE.md / PLAN.mdCurrent condition or pending work must be externalized
SNAPSHOT.mdA point-in-time state record is useful for handoff or audit
Prompt filesThe agent’s startup path requires structured prompt input
Tool registryTool definitions, permissions, or schemas need a central reference
Continuity artifacts such as durable memory or heartbeat instructions may exist in some systems, but this template does not prescribe them as normative outputs.

​
What it configures

  • Agent instructions (scope, role, boundaries, behavior rules).
  • Model or provider settings when applicable.
  • Tool definitions, permissions, and invocation rules.
  • Prompt files and prompt references.
  • Workspace directories.
  • State or context references.
  • Expected inputs and outputs.

​
The procedure

1

INSPECT

Identify the agent’s intended role and the target it operates on.
  • Inspect existing AGENTS.md files, prompts, tool definitions, provider or model configuration, workspace directories, state files, context files, schedules, expected inputs and outputs, permissions, and operator constraints.
  • Preserve valid existing instructions. Respect nested instruction scopes when they exist.
  • Use existing agent files, project docs, tool schemas, prompt files, configuration, and user instructions as sources.
2

DETERMINE

Decide what the agent needs to become operational in this target.
  • Identify required instructions, tool access, prompt structure, provider configuration, state representation, context handoff, validation checks, and boundaries.
  • Produce a target-specific agent initialization plan.
Possible decisions include:
  • Create AGENTS.md only when agent operation applies and no valid instruction file exists.
  • Create SELF.md only when identity, role, capabilities, or boundaries need to be explicit.
  • Create USER.md only when operator context materially affects operation.
  • Create TOOLS.md only when tool permissions or invocation rules need durable documentation.
  • Create STATE.md or PLAN.md only when current condition or pending work must be externalized.
  • Create SNAPSHOT.md only when a point-in-time state record is useful.
3

CREATE

Create only applicable artifacts from the conditional list above.
  • Every artifact must be derived from the target. Do not copy generic instruction files without adaptation.
  • Preserve and repair existing valid instructions rather than replacing them.
4

CONFIGURE

Make the agent artifacts usable together.
  • Configure instruction precedence, model or provider settings, tool permissions, prompt references, state locations, input and output conventions, schedules when applicable, and validation commands.
  • Configure only what is required for this agent target.
5

VALIDATE

Confirm the agent can operate with the initialized instructions and tools.
  • Check instruction consistency, required files, tool availability, provider configuration, prompt references, expected input and output paths, state handoff, and any startup or dry-run command the target supports.
  • Report status as OPERATIONAL, OPERATIONAL_WITH_WARNINGS, BLOCKED, or DRY_RUN_COMPLETE.
6

REPORT

Report the complete agent initialization result:
  • Agent target identified.
  • Role, boundaries, and operating context.
  • Inspected instructions, prompts, tools, and state.
  • Created or updated artifacts.
  • Configuration performed.
  • Validation result.
  • Warnings, missing access, or blockers.
  • Final status.

​
Agent-specific notes

  • Different targets need different instructions. An AGENTS.md written for one repository is not valid for another. Instructions must reflect the specific target’s structure, tools, and conventions.
  • Nested projects may require nested instruction files. When a monorepo contains sub-packages that agents operate on independently, each sub-package may need its own scoped AGENTS.md.
  • Tool access must reflect real permissions. Do not document tools the agent cannot actually access. Do not omit tools the agent requires. The tool definitions created or referenced in this template must match the runtime environment.
  • Agent readiness is validated by the target’s startup path. The validation step uses the specific checks the target supports — not a generic checklist applied to every agent.

​
Using this template

Initialize this agent workspace using the agent init template.
Inspect existing instructions, tools, and state files.
Create only the artifacts this agent target actually requires.
Validate that the agent can operate with the initialized configuration.

Build docs developers (and LLMs) love

Get started for freeTalk to us