omp packages

pi-session-inspector0.1.0

Branch-aware, bounded Pi session inspection tool

boadij Published 1 month ago ~193 downloads/mo

Untested on omp

This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.

Did it work for you?

Result

One report per person. A new report replaces your old one.

Reports 0

No reports yet.

Readme

Pi Session Inspector

pi-session-inspector adds one model-callable Pi extension tool, session_inspect, for bounded, branch-aware inspection of the active session or an explicitly supplied historical .jsonl session.

Install

pi install /absolute/path/to/pi-session-inspector

Reload Pi after installing.

Tests

With Pi 0.85.1 installed and available as pi, run the regression suite through Pi's bundled extension loader:

bun run test

The test command does not install or bundle Pi's peer packages.

Actions

session_inspect supports:

  • summary - session metadata, message/tool counts, errors, token/cost usage, compactions, leaves
  • timeline - bounded compact entry timeline
  • tools - correlated tool calls/results with status
  • errors - tool and assistant errors
  • search - case-insensitive targeted session search
  • branches - leaf/branch overview
  • entry - exact bounded entry inspection

Scopes:

  • branch - current active path, default
  • all - entire session tree
  • context - active branch after Pi's context/compaction rules

By default the tool inspects the active session. session may point explicitly to another Pi .jsonl session file; historical files are parsed and migrated in memory and are never rewritten.

Typical agent workflow

session_inspect(action="summary")
session_inspect(action="errors", limit=10)
session_inspect(action="search", query="src/auth.ts", limit=10)
session_inspect(action="entry", entryId="...")

The tool excludes its own session_inspect calls from tool-call statistics and its tool results from tool-result statistics; the assistant messages and usage that produced them remain normal session data. Output is bounded, redacted, and omits thinking blocks and image payloads.