Interaction Protocols

Conventions that define how the system speaks to itself. Each symbol signals a different intent to an agent or the system.

#Symbol Reference

Symbol	Function	Example	Semantics
#path	Load context (discovery mode)	#inscription	Navigate/explore this topic, answer questions about it
&path	Load context for design (architecture mode)	&inscription	Load this topic AND help me design/architect it
@agent-name	Address an agent directly	@claude-code-guide	Send message to a specific agent or person
/skill-name	Invoke a skill	/write	Execute a skill with its specialized workflow

#Discovery Mode (#path)

Signal: Load context for exploration and Q&A.

What happens:

1. Agent loads the specified path (project, folder, document, or skill)
2. Agent reads the relevant index.md and related files
3. Agent is ready to answer questions, explain structure, or provide navigation
4. Agent stays in reading/explanation mode

Examples:

• #inscription → Load inscription project context, then answer questions about it
• #inscription/theorem 5 → Navigate to theorem-5 subthread, then explain it
• #face-states → Load face-decoding project context
• #inscription can we revisit theorem 5? → Load context, then answer the question

When to use: Exploration, understanding, learning, asking questions about existing work.

#Architecture Mode (&path)

Signal: Load context AND engage architecture/design-level reasoning.

What happens:

1. Agent loads the specified path (project, folder, or document)
2. Agent reads structure, dependencies, existing patterns
3. Agent engages the Plan agent (if complexity warrants) or architecture-level reasoning
4. Agent proposes design strategies, identifies critical paths, flags dependencies
5. Agent is ready to help you design or redesign the thing

Examples:

• &inscription → Load and help me architect the inscription project
• &inscription/theorem 5 → Load and help me redesign theorem 5
• &federation — how would you structure this? → Load and propose architecture
• &projects/ai-ontology — what's the critical path? → Load and analyze dependencies

When to use: Design decisions, refactoring, architectural review, critical path analysis, dependency mapping, "help me think through this."

#Agent Address (@agent-name)

Signal: Send a message to a specific agent or person.

What happens:

1. Agent looks up the addressee (human or AI agent)
2. Message is routed directly to that agent
3. Used for explicit handoff, cross-agent communication, or person mention

Examples:

• @claude-code-guide — where is the MCP docs? → Ask the guide skill
• @zephyr — here's the Paper 6 outline → Message Zephyr directly
• @plan-agent — can you design the critical path? → Engage Plan agent

When to use: Direct requests to specific agents, cross-agent coordination, handoff notation.

#Skill Invocation (/skill-name)

Signal: Execute a skill's specialized workflow.

What happens:

1. Skill is invoked with its full procedure (Define → Collect → Reason → Act → Validate, or similar)
2. Skill may ask clarifying questions
3. Skill performs its function autonomously
4. Skill returns results or artifacts

Examples:

• /write inscription/theorem 5 → Invoke write skill for theorem 5
• /read inscription → Invoke read skill for discovery
• /commit — add theorem 5 proof → Invoke commit skill with message
• /board add — Paper 3 draft complete → Invoke board skill to post update

When to use: When you want a skill's full procedure, not just a manual action.

#Combining Symbols

Load context + ask question:

#inscription — can we revisit theorem 5?

The # prefix ends at the first punctuation mark or conjunction. Symbol loads context, then the question is answered.

Load for design + ask question:

&inscription — what's the critical path?

Symbol signals architecture mode, then the question is addressed with design-level reasoning.

Address agent + invoke skill:

@plan-agent, can you /write — a critical path analysis for inscription?

Agent is addressed; they are asked to use a skill.

#Precedence & Ordering

1. First symbol determines mode: #, &, @, or / at the start of a message determines how the message is routed
2. Question/clause comes after: After symbol and path, punctuation marks the boundary before a question or instruction
3. Multi-symbol messages: Can combine (e.g., address agent to invoke skill), but the first symbol always determines primary mode

---

haak · created 2026-02-28 · zach + claude

Methods 17 — Interaction Protocols — 2026 — Zachary F. Mainen / HAAK