[FEATURE]:add ability to run commands after every chat ends with changed files

[alosslessdev]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

I want opencode to be able to run custom shell commands Every time a chat ends with files modified by the ai agent. This can be running git commit and git pull and push as well as running gitnexus index. This would automate git commits every time the repo is modified by an ai agent to allow for easy rollbacks

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [FEATURE]: add session.turn.completed bus event for plugin hooks #23503: Directly overlaps — proposes a session.turn.completed hook, explicitly listing "auto-snapshot plugins (git commit after each turn)" as a motivating use case, which matches this request
• Native Claude Code hooks compatibility (PreToolUse, PostToolUse, Stop) #12472: Related — covers native Claude Code hooks parity (PostToolUse, Stop), which provides similar post-turn lifecycle hooks
• [FEATURE]: add session.stopping plugin hook to allow re-entering the agent loop #16626: Related — proposes a session.stopping plugin hook to allow re-entering the agent loop at session end
• [FEATURE]: SessionStart hook for session lifecycle events #5409: Related — proposes a SessionStart hook for session lifecycle events (complementary lifecycle hook)

[alosslessdev]

I think it's similar to the first 2 above but it directly runs bash commands instead. No plugins or installation are required on opencode. Opencode would directly execute bash commands without agent input, if it detects files modified by the agent during the chat session

Of course it can be quite dangerous. To mitigate this it can maybe run as another user via ssh within the same machine, which is then granted access to the working folder via Linux permissions

It could work with hooks that are not running bash commands to inject text into for example commit messages, like spawning a subagent to summarize changes

[Keesan12]

The useful boundary here might be a post-turn hook that receives a small receipt first, not an unconditional "run arbitrary bash after every changed session" path.

Something like:

{ dirty_files, changed_paths, branch, exit_reason, tests_run, approval_state }

Then users can wire commit/index/summarize flows however they want, while the core can keep obviously risky actions like git push or networked side effects behind explicit policy.

That would still cover the auto-commit / gitnexus indexing use case, but make the safety boundary easier to reason about. If you want, I'm happy to help pressure-test a minimal hook contract that supports this workflow without normalizing surprise pushes.