entire-query

eq — fetch Entire AI session history from a Git repo as stable JSON / JSONL. Like jq for Entire checkpoints.

eq complements the official entire CLI. Use entire for human-facing workflows such as setup, semantic search, quick explanations, summaries, rewind, and resume. Use eq when agents, skills, or scripts need reproducible structured output for citations, token aggregation, transcript audits, or other deterministic workflows.

The npm package is entire-query; the binary is eq.

Agent and skill use

eq is designed primarily as an agent-facing tool, especially when paired with skills, prompts, or other reusable agent instructions. Humans can run it directly, but its main job is to give agents stable, scriptable access to Entire checkpoint data without relying on human-oriented terminal views.

To install the companion skill for an agent environment:

npx skills add morinokami/entire-query

A typical skill should call eq to retrieve primary session data, then do any summarization, comparison, citation, or filtering in the agent layer. Prefer eq over entire explain when the agent needs deterministic JSON/JSONL, exact prompts, transcript events, token usage, or checkpoint metadata.

Run

npx -y entire-query <subcommand> [...args]

No install required. Optional global install: npm i -g entire-query.

Commands

eq checkpoint <id-or-ref>                   # one Checkpoint JSON; git refs resolve via Entire-Checkpoint trailer
eq checkpoint list [--file <path>]          # NDJSON stream
eq session list <id> [--file <path>]        # NDJSON of SessionSummary (with prompt_preview)
eq session get <id> --index <n>             # one Session JSON
eq prompt <id> --session <n>                # session's prompt.txt as JSON
eq transcript <id> --session <n> [--role <user|assistant|tool|unknown>]   # NDJSON of TranscriptEvent

All commands accept --repo <path> (default: cwd). JSON commands auto-emit compact JSON under AI agents; pass --json explicitly otherwise. Errors are {"error":{"kind":"...","message":"..."}} on stderr with non-zero exit.

Why

entire explain is useful for getting oriented and finding likely checkpoints. It is a human-readable CLI view: it can show a stored AI summary, entire explain --generate creates one, and the default view also renders metadata, prompts, and parsed transcript excerpts directly. eq is for the next step: reading the underlying checkpoint data as raw structured output so callers can quote, aggregate, audit, and cross-reference primary sources. Aggregation is delegated to jq.

eq currently reads the entire/checkpoints/v1 branch. For checkpoints v2-only repositories or workflows that need remote fetch, generated summaries, rewind, or resume, use the official entire CLI.