Entire enabled but commits never get Entire-Checkpoint trailer (hooks run, no linkage)

[dayhaysoos]

What happened?

Hello! I'm seeing a reproducible issue where Entire is enabled andhooks are installed, but new commits do not receive Entire-Checkpoint trailers.

Steps to reproduce

1. Make sure Entire is enabled (I'm on the latest version 0.5.1)

entire enable --force --absolute-git-hook-path --agent opencode
entire status
ls -l .git/hooks/prepare-commit-msg .git/hooks/commit-msg .git/hooks/post-commit

2. Stage any file change
3. Commit with debug:

ENTIRE_LOG_LEVEL=DEBUG git commit -m "checkpoint debug test"

4. Check commit linkage

git show -s --format=%B HEAD
entire explain --commit HEAD --no-pager
entire trace

What happens:

• Commit succeeds, but commit message has no trailer:
• no Entire-Checkpoint: ...
• entire explain --commit HEAD --no-pager returns:
  • No associated Entire checkpoint
  • Commit does not have an Entire-Checkpoint trailer.
• entire trace only shows post-commit:
  • post-commit ...
  • step: open_repository_and_head
    entire doctor reports metadata branches OK (no obvious repair action suggested).

Entire CLI version

Entire CLI 0.5.1 (d46fdc2)

OS and architecture

Darwin 25.3.0 arm64

Agent

GPT-5.3 Codex

Terminal

OpenCode Desktop terminal (I suppose it's Ghostty?)

Logs / debug output

# Config confirms Entire is enabled with always-linking
.entire/settings.json:1-6
{
  "enabled": true,
  "absolute_git_hook_path": true,
  "telemetry": true,
  "commit_linking": "always"
}


# Commit attempt 2 (same result)
.entire/logs/entire.log:6329-6374
{"time":"2026-03-24T23:24:34.756062-04:00","level":"DEBUG","msg":"prepare-commit-msg hook invoked","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"hooks","hook":"prepare-commit-msg","hook_type":"git","strategy":"manual-commit","source":"message"}
{"time":"2026-03-24T23:24:34.780429-04:00","level":"DEBUG","msg":"stagedFilesOverlapWithContent: no overlapping files found","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"checkpoint","staged_files":1,"files_touched":1}
{"time":"2026-03-24T23:24:34.780441-04:00","level":"DEBUG","msg":"sessionHasNewContent: no transcript, carry-forward with staged files","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"manual-commit","session_id":"ses_2f84f8dc1ffeWdjA1lTVXOJuGQ","files_touched":1,"staged_files":1,"result":false}
{"time":"2026-03-24T23:24:34.780726-04:00","level":"DEBUG","msg":"prepare-commit-msg: no content to link","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"checkpoint","strategy":"manual-commit","source":"message","sessions_found":38}
{"time":"2026-03-24T23:24:34.807891-04:00","level":"DEBUG","msg":"commit-msg hook invoked","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"hooks","hook":"commit-msg","hook_type":"git","strategy":"manual-commit"}
{"time":"2026-03-24T23:24:34.844439-04:00","level":"DEBUG","msg":"post-commit hook invoked","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"hooks","hook":"post-commit","hook_type":"git","strategy":"manual-commit"}
{"time":"2026-03-24T23:24:34.849628-04:00","level":"DEBUG","msg":"perf","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"perf","op":"post-commit","duration_ms":5,"steps.open_repository_and_head_ms":0,"hook_type":"git"}


# Additional context: transcript export repeatedly failing around same period
.entire/logs/entire.log:6091-6092
{"time":"2026-03-24T08:23:22.29529-04:00","level":"WARN","msg":"failed to prepare transcript","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"lifecycle","agent":"opencode","error":"opencode export returned invalid JSON (66048 bytes)"}
{"time":"2026-03-24T08:23:23.354505-04:00","level":"INFO","msg":"checkpoint skipped (no changes)","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"checkpoint","agent":"opencode","strategy":"manual-commit","checkpoint_type":"session","checkpoint_count":6,"shadow_branch":"entire/69f1d19-e3b0c4"}

Additional context

One thing I want to mention that might have been the catalyst. When I originally set up Entire, I did the quick start command and selected OpenCode at the time.

Today, I was experiencing in consistencies so I figured I should run entire enable again with the OpenCode flags assuming maybe something will get fixed. However, since then, I haven't been able to capture Checkpoints at all. Not sure if this could have caused it but figured it was worth sharing.

Let me know if you have any questions!

UPDATE!!

It looks like my agent can create Checkpoints but I can't. This was the reply. Still troubleshooting.

I created an empty commit: 1c512f9 with message agent checkpoint canary.
The commit did get a trailer:
Entire-Checkpoint: 1c9dfafc4479
But entire explain --commit HEAD --no-pager fails with:
checkpoint not found: 1c9dfafc4479
So checkpoint injection is happening now, but checkpoint lookup/storage is inconsistent (trailer points to a checkpoint ID Entire can’t resolve). That’s a new, very useful signal for your Entire issue.```

[pfleidi]

Thanks for the detailed report and logs, that's super helpful!

Based on the logs you've sent, it looks like invoking the opencode export command returned invalid JSON. I'm referring to this line:

{"time":"2026-03-24T08:23:22.29529-04:00","level":"WARN","msg":"failed to prepare transcript","session_id":"ses_2e7798cf4ffe51qGR1tS3y3sM7","component":"lifecycle","agent":"opencode","error":"opencode export returned invalid JSON (66048 bytes)"}

Can you do us a favor and help to understand what this command returns?

1. Find your active session ID

ls .git/entire-sessions/

Pick the most recent .json file — the session ID is the filename without .json.

2. Check if opencode export produces valid JSON

If you have jq installed, this is probably the simplest way to do it:

opencode export <session-id> | jq

Otherwise, this should do:

opencode export <session-id> > /tmp/entire-export.txt 2>/tmp/entire-export-err.txt

# Check sizes
wc -c /tmp/entire-export.txt /tmp/entire-export-err.txt

# Check if it's valid JSON
python3 -c "import json; json.load(open('/tmp/entire-export.txt')); print('Valid JSON')"

# If invalid, can you show what's at the start and end of the output?
head -5 /tmp/entire-export.txt
tail -5 /tmp/entire-export.txt

3. Dump current session states

for f in .git/entire-sessions/*.json; do echo "=== $(basename $f) ==="; cat "$f" | python3 -m json.tool; done > /tmp/entire-sessions.txt

This might contain many sessions — feel free to attach /tmp/entire-sessions.txt as a file.

[dayhaysoos]

Hey @pfleidi, thanks for responding! I made test branches to reproduce the issue so the sessions are really small.

entire-sessions.txt

{
  "session_id": "test123",
  "cli_version": "0.5.1",
  "base_commit": "942eba6720781dcf27ba8565563d886db4fbb25b",
  "attribution_base_commit": "17f1ca86f137138d53ffc297bab50d11bb4c1af9",
  "worktree_path": "/Users/nickdejesus/Code/nimbus",
  "started_at": "2026-03-25T09:42:18.717691-04:00",
  "phase": "active",
  "turn_id": "60abe9dcc20a",
  "last_interaction_time": "2026-03-25T09:42:18.718173-04:00",
  "checkpoint_count": 0,
  "agent_type": "OpenCode",
  "model_name": "test",
  "transcript_path": "/Users/nickdejesus/Code/nimbus/.entire/tmp/test123.json",
  "last_prompt": "test",
  "pending_prompt_attribution": {
    "checkpoint_number": 1,
    "user_lines_added": 0,
    "user_lines_removed": 0,
    "agent_lines_added": 0,
    "agent_lines_removed": 0
  }
}

The error version:

Exporting session: test123
schema validation issues: [
  {
    "origin": "string",
    "code": "invalid_format",
    "format": "starts_with",
    "prefix": "ses",
    "path": [],
    "message": "Invalid string: must start with \"ses\""
  }
]
�[91m�[1mError: �[0mSession not found: test123

[pfleidi]

@dayhaysoos Is this the actual session ID? Based on the output returned from opencode exports it looks like the session ID doesn't exist.

Opencode should also show the session ID after quitting the agent, so you can resume again:

❯ opencode
                                   ▄
  █▀▀█ █▀▀█ █▀▀█ █▀▀▄ █▀▀▀ █▀▀█ █▀▀█ █▀▀█
  █  █ █  █ █▀▀▀ █  █ █    █  █ █  █ █▀▀▀
  ▀▀▀▀ █▀▀▀ ▀▀▀▀ ▀▀▀▀ ▀▀▀▀ ▀▀▀▀ ▀▀▀▀ ▀▀▀▀

  Session   Greeting message
  Continue  opencode -s ses_<rest of session id>

[dayhaysoos]

@pfleidi my mistake! I actually use OpenCode Desktop so I never saw the IDs. I believe THIS is what you're looking for:

entire-export.txt

This was a massive session, I normally would keep them smaller but I promise you it was in the name of science. Let me know if you need anything else!

[pfleidi]

@dayhaysoos The OpenCode desktop app is quite new. We need to do more testing with it. Did you, by any chance, try the opencode CLI version? If so, did that work as expected or did you have the same issues as with the desktop app.

In the meantime, I've opened a PR to add more logging to make diagnosing such issues easier in the future: #785

[dayhaysoos]

@pfleidi I can try with the CLI version. If you don't mind, I think I want to take a crack at solving this myself. I'll spend some time exploring and if I can't solve it, I'll try the CLI version and see if it works.

[pfleidi]

@dayhaysoos Of course!

[dayhaysoos]

@pfleidi Hey, just an update. I switched to the CLI and it didn't yield and results. I even tried wiping my installation of Entire and reinstalled and it still didn't work.

HOWEVER

I have been trying to solve this Entire Checkpoint issue within the same Agent session on OpenCode. I switched branches and stuff but I always persisted the same session. I decided to try making a new session and now all of a sudden it works.

I'm suspecting that something really funky happened within that session that doesn't allow Entire to create a Checkpoint, despite an Entire Trailer existing. I'm going to keep my eye out for things as I move forward, but I might be okay for now. I don't know if you want to close this issue or what, but figured I'd let you know how my exploration went

[dayhaysoos]

I think I've identified exactly where things are going wrong! So I just tried doing a soft reset because I wanted to ensure I was making a checkpoint for the commit I had going, but this seems to ruin the meta data needed to generate the checkpoints. I think this may be a bug because this worked just fine in 0.49 (and was actually suggested by staff in the discord when I was having issues). Now it seems like running git reset guarantees metadata going stale.

I could be wrong about this but I think that's what's going on