@fradser/pi-context

@fradser/pi-context exposes one tool: context_get.

It starts an independent Pi child process for repository, library, codebase, and technical-topic research. This isolates exploratory tool calls from the main session while returning a concise, evidence-based result.

Installation

pi install npm:@fradser/pi-context

Usage

Use natural language to ask Pi for research or external context:

Research how facebook/react implements server actions.
Compare the current React Router and Next.js data-loading guidance.
Investigate https://github.com/owner/repo and explain its architecture.

Pi recognizes these requests and invokes context_get automatically. Users do not need to type context_get or use a slash command directly; it is an internal native tool with one query parameter containing the context request.

Child-process constraints

The child runs as:

pi --print --mode json --no-session -ne -ns -np -nc --no-themes --tools read,bash

The child runs in the caller's working directory through pi-kit's minimal runPiWorker mode: -ne -ns -np -nc --no-themes disables extension, skill, prompt-template, context-file, and theme discovery, while --tools read,bash exposes only the two required tools. It has no sandbox, timeout, or display truncation; bash remains technically able to write because there is no OS sandbox, so the bundled worker prompt requires no modifications. If line-level evidence from a public repository is necessary, it may use git clone --depth=1 under /tmp with removal after inspection. It must not run package-management, deployment, or interactive commands. The shared worker fails closed if its bounded stdout, stderr, or JSONL line limits are exceeded.

A package-local typed builder reads the bundled Markdown as a reference protocol and builds a task-specific context-review prompt containing the caller working directory, complete research request, reference protocol, and completion contract. The Markdown file itself is never used as an execution identity. Startup renders [context] research started · <research query>, using the normalized concrete query that produced the prompt; the row remains width-bounded and does not expose the resource path. While the async child runs, the transcript keeps only that started row with no duplicate progress row; live progress shows in the researcher widget above the editor. The widget shows neutral live research activity. The child always runs with --no-session, so separate invocations retain no memory or agent-teams state. A successful empty answer is retried exactly once with a stronger completion requirement that does not rely on hidden reasoning or prior tool output. Results enter the main session untruncated and render as a compact, expandable [context] researched lifecycle row.

Structure

context/
├── index.ts
├── extensions/context-tools.ts
├── extensions/context-command.ts
├── references/workflow.md
├── extensions/context-prompt.ts
└── prompts/context-research.md

License

MIT