Agent interfaces

Role

Lets short commands become sourced, validated actions while the vault remains the source of truth.

Contains

• Hermes command bus
• Skills as procedural memory
• Telegram text/voice capture
• Repo/web/calendar/email tools that write back with provenance

Interfaces

• Reads vault context, acts through tools, writes back via workflow rules
• Routes reusable work through /skills/
• Keeps session/memory/fact stores as cache unless mirrored

Maintenance rule

Interfaces may be fast and conversational, but durable changes still need provenance and vault writeback.

Reliability signal

A command can be traced from Telegram/session input to vault context, repo output, or an explicit blocker.

What goes in

Agents should receive bounded context, clear read/write permissions, and the exact source path or workflow contract that governs the operation.

What comes out

A good agent output leaves traceable evidence: changed files, validation results, provenance, and a handoff that another agent or human can resume.

Quality check

Did the agent receive only bounded context, did it respect read/write scope, and did the output include evidence instead of a vague done-claim?

Concrete example

A Telegram instruction that changes a repo should resolve to a bounded task, inspect source context, edit only allowed files, run mechanical checks, and return commit plus verification evidence.