Search protocol

Defines how to search across information sources for answering queries, analysis, or communication.

Information hierarchy

memory/

journal/

contexts/

Index-first pattern (MANDATORY)

Every search reads an

_index.md

• memory/_index.md

  routes to the correct subfolder

• memory/{category}/_index.md

  maps aliases to filenames

• journal/YYYY-MM/_index.md

  lists entries by date, type, participants, initiatives

Never scan all files in a folder. Always use the index.

When to search each source

Memory

People, initiatives, products, vendors, competitors, past decisions.

Tasks

What needs to be done, deadlines, assignments, follow-ups.

Journal

Recent discussions, who said what, decisions from meetings, prior briefings.

• Optimization: For "last week" queries, read only current month's

  _index.md

• Optimization: For "find all meetings about X", check recent

  _index.md

  files first
• Optimization: For "what did we decide about X", check

  memory/decisions/

  first

Contexts

Deep detail beyond memory: full product docs, schemas, org charts, specs.

Calendar (via MCP or legacy integration)

Priority flow:

1. Check

   <mcp_servers>

   context for calendar MCP server (preferred)
2. If found, use MCP tools (

   list_events

   ,

   get_event

   ,

   create_event

   )
3. If not found, check

   reference/integrations.md

   for legacy provider (eventlink)
4. If legacy configured, execute HTTP API calls
5. If neither exists, note unavailable and suggest: "Run /welcome to configure calendar integration"

Always resolve dates to YYYY-MM-DD before querying.

Tasks (via MCP or legacy integration)

Priority flow:

1. Check

   <mcp_servers>

   context for reminders/tasks MCP server (preferred)
2. If found, use MCP tools (

   list_reminders

   ,

   create_reminder

   ,

   complete_reminder

   )
3. If not found, check

   reference/integrations.md

   for legacy provider (remindctl)
4. If legacy configured, execute CLI commands
5. If neither exists, fall back to workspace file data (Active/Delegated/Backlog folders)

MCP tools

• ~~project tracker: Issue status, sprint data, initiative metrics
• ~~documentation: Documentation, policies, technical specs

Web

Market information, competitor news, regulatory updates, current events. External only.

Search procedure

Step 1: Parse the query

Identify:

• Entities: People, products, initiatives, vendors, competitors
• Topics: What the query is about
• Timeframes: Dates, deadlines, recency
• Depth: Summary or full detail needed?

Step 2: Check aliases

Entities may have alternate names. Check

_index.md

Step 3: Search memory (index-first)

1. Read relevant

   _index.md

   to find target file
2. Read the specific file
3. Follow

   [[Entity]]

   wikilinks for related context

Step 4: Check tasks (if applicable)

Execute the task integration

list

Step 5: Check journal (if applicable)

1. Read

   journal/YYYY-MM/_index.md

   for recent months
2. Open specific entries matching the query
3. Expand backward in time only if needed

Step 6: Check deep contexts (if needed)

Look in

contexts/

Step 7: Cite sources

When using information from searches, cite the source:

[From memory/people/jane-smith.md]
Jane prefers data-driven summaries and email over Slack.

[From journal/2026-01/2026-01-18-planning-session.md]
Discussed moving the deadline to end of February.

Quick answer mode

For fast factual lookups:

Scheduling queries (calendar, agenda, meetings, availability, "am I free")

Start with the calendar integration FIRST. Read

reference/integrations.md

YYYY-MM-DD

list_events

All other queries

Follow standard hierarchy: memory -> tasks -> journal -> contexts -> MCP -> web.

Gap closure

After searching, apply clarification protocol if critical context is still missing.

Context budget

• Memory: Read

  _index.md

  + up to 5 targeted files
• Journal: Current month

  _index.md

  + up to 3 entries
• Tasks: Execute task integration

  list

  operation for Active only (unless other lists explicitly needed)

Source attribution

When answering, tag each piece of information with its confidence tier:

Absolute constraints

Universal constraints from the core skill apply (date resolution, integration constraints, index-first pattern, wikilink mandate). Additionally:

• NEVER answer internal questions from web search alone
• NEVER hallucinate memory that doesn't exist
• NEVER skip context search when deep detail is clearly needed
• NEVER claim calendar access is unavailable. TARS has calendar access via configured integration. If integration fails, report the specific error.
• ALWAYS query calendar integration for any question about schedule, agenda, meetings, availability, or "am I free"
• ALWAYS check aliases when entity not found by primary name