---
name: inferenceos-context-briefing
description: Generate a compact workspace briefing via InferenceOS to start sessions with near-zero tokens instead of replaying full context.
user-invocable: true
---

# InferenceOS Context Briefing

## Objective
Generate a compressed workspace snapshot (file counts, language breakdown, symbol counts, changed files, key file signatures) so Claude can orient to a codebase without reading dozens of files. This saves significant tokens on session starts.

## Execution Workflow
1. Call the `inferenceos_route_task` MCP tool with the user's task to get a pre-compressed context bundle and repo snapshot.
2. Review the `context_bundle.compact_brief` and `repo_context` from the route result — this is the workspace briefing.
3. Use the briefing to orient before reading any files. Only read files that the briefing flags as relevant.
4. If the briefing shows changed files, prioritize those over full workspace scans.
5. After completing work, call `inferenceos_persist_result` to cache the successful route for future reuse.
6. Call `inferenceos_auto_store_turn` to persist the conversation turn locally so the next session can recall it.

## When to Use
- At the start of any coding session, before doing broad file reads
- When switching between tasks in the same workspace
- When context is getting large and you need to re-orient

## Token Savings
Instead of reading 10-20 files to understand a codebase (~20K tokens), the briefing gives you the same orientation in ~500 tokens. Use it first, then read only what's needed.

## Red Flags & Constraints
- Do not skip the briefing and jump to reading files — the whole point is to orient cheaply first.
- Do not ignore changed files in the briefing — they are the most likely relevant context.
- Always persist results after completing work so future sessions benefit.