Versioning

Reading history

agit log — turn by turn

agit log                  # current session's branch
agit log exp              # another branch
agit log alice/photo@exp  # someone else's

One line per turn commit: turn number, short SHA, message (the prompt’s first 72 chars). Milestones stand out. Use -v/--context to see events inside a turn.

agit show — one point, readable

agit show exp#12      # turn 12 rendered as a conversation
agit show exp#12.3    # just event 3 of turn 12 (a tool call, a file write…)
agit show exp:path    # a file as of that point

agit diff — what changed between two points

agit diff exp#10 exp#14

Shows new turns between the two, plus shared-file changes (memory/, skills/).

agit view — what the session actually sees

agit view exp          # VIEW composition, event by event
agit view exp --json   # for tooling

The VIEW is what gets materialized when a session resumes. Marker and summary envelopes (from merges) are part of it. If you want to know “what will the agent see when it wakes up”, this is the answer.

Reference syntax, all of it

Shape	Meaning
exp, agit-91c7faef	branch head / commit SHA
exp#12 · exp#-1	turn 12 · latest turn
exp#12.3	event 3 inside turn 12
exp#4..#9	turns 4 through 9
owner/repo@exp	any published line
exp:path	file content at that point
@	your current session
@exp	your session, arbitrated onto branch exp

You can mix: alice/photo@exp#12.3 is a legal address.

agit search

agit search "rate limit retry"            # has anyone done this?
agit search "表3" --repo alice/photo

Searches the corpus you can read (public agents + your privates). The primary entry point is MCP — an agent asks mid-run without leaving its runtime.

agit export

agit export exp --format markdown -o out.md
agit export exp#4..#9 --format jsonl

Formats: jsonl (raw envelopes), ir, markdown, claude-code, codex. --view-only exports only what the session sees; --redact masks secrets on the way out.