edhil log

edhil log [NUMBER] [--epic NUMBER] [--work-item NUMBER] [-n COUNT]
          [--item NUMBER [--step NUMBER] [--patch]]

Shows what the runs of this repository did. It reads only the journal in .edhil/journal.db; it needs no configuration and no token, and it can run while another terminal is running edhil implement.

The list of runs

Without a number the runs are listed, newest first, one line each: the run number, when it started, its state, what it was about and how many work items were committed.

#12  2026-10-27 09:14  done     epic #5 The epic        3/3 committed
#11  2026-10-26 16:02  aborted  work items #30 #31      1/2 committed, 1 paused
#10  2026-10-26 11:40  stopped  work item #22           0/1 committed

The states are:

State Meaning
done Every work item of the run was committed.
aborted The run was ended at a pause, or the process died.
stopped The run was stopped with Ctrl-C.
running The run is in progress, perhaps in another terminal.
paused The run waits for an answer.
Option Meaning
-n, --limit COUNT How many runs to list. The default is 20. When there are more, stderr says how many were left out.
--epic NUMBER Only the runs of this epic.
--work-item NUMBER Only the runs that included this work item. A run of several epics or of loose work items has no epic of its own, so this is how you find it.

Without any run it prints No runs yet.; when the filters match nothing it prints No runs match.

One run

edhil log 12

prints the report of run 12 as markdown: the state, the branch, the agent that did the work (its name, its tool and its model, such as strong (claude, model sonnet)), the times and a table of its work items with their phases and commits. A run made before agents existed has no agent line.

# Run 12 - Epic #5: The epic

- Branch: main
- Base commit: 0123456789abcdef0123456789abcdef01234567
- Agent: strong (claude, model sonnet)
- Started: 2026-10-27T09:14:02.000Z
- Ended: 2026-10-27T09:41:10.000Z (27m 8s)
- Final state: done
- edhil version: 0.1.0

## Work items (3 of 3 committed)

| # | Work item   | Status    | Duration | Commit  |
|---|-------------|-----------|----------|---------|
| 1 | #11 First   | committed | 9m 2s    | 0123456 |
| 2 | #12 Second  | committed | 8m 40s   | 89abcde |
| 3 | #13 Loose   | committed | 9m 26s   | fedcba9 |

The tables are markdown with the columns padded, so they read well in the terminal and in a markdown viewer.

One work item of a run

edhil log 12 --item 123
edhil log 12 --item 123 --patch

prints the report of one work item: its steps, numbered within the work item (implement, verify, fix, summarize), with the command of every verify step in a Command column, its events (pauses and the answers) and its commit message. With --patch it also prints the patch of the commit, which the journal keeps, so it is available even if the commit has been rebased away.

One step

edhil log 12 --item 123 --step 2

prints one step: how it ended, the prompt the agent was given, its transcript and everything the step printed, line by line with the time and the stream (out or err). For a verify step it also names the command that was run. This is where to look when an agent did something strange or a verification failed.

Good to know

  • The numbers are run numbers here, not issue numbers, except for --epic, --work-item and --item, which are issue numbers.
  • Long output goes through the pager ($PAGER, less -FRX by default) when stdout is a terminal. --no-pager and pipes get it directly.
  • A number that does not exist is an error with exit code 1 and a message that says which run or work item was not found.