context-mode

Files (1)

• SKILL.md

SKILL.md (first lines)

# Context Mode

Use Context Mode when an operation can return more than about 20 lines. Also use it when you must process the full input but need only a compact result in the conversation

Good targets include:
- builds, tests, logs, dependency reports, and broad Git output
- large JSON, API responses, datasets, and documents
- browser snapshots, console messages, and network requests
- broad MCP results that need filtering, comparison, or synthesis

Keep normal Pi tools for small file reads, narrow searches, source navigation, and short command output

## Choose the path

Use the `context-mode` MCP server through the `mcp` gateway:
- Use `context_mode_ctx_execute` when a command, fetch, or script can process the data directly in the sandbox
- Use `context_mode_ctx_execute_file` when the data already exists in a file
- Use `context_mode_ctx_index` with a file path when the source needs repeated search, then use `context_mode_ctx_search` for focused retrieval

Print only the result needed for the current decision. Do not truncate the input before analysis merely to reduce output

## File-first flow

When an action or another tool can save large output directly, save it in a file first. Pass that path to `context_mode_ctx_execute_file` or `context_mode_ctx_index`. Do not first return the result and then send the same content again through an inline `content` argument

Use the same file-first flow for large console output, network requests, generated reports, and local data

From OrestesK/pi, MIT.

More from OrestesK/pi