feat: auto-fix-lint hook with follow-up prompts #56

New issue

Closed

opened 2026-05-28 11:43:02 +02:00 by dries · 0 comments

dries commented

2026-05-28 11:43:02 +02:00

Owner

From spec/multi-agent-support/agent-feature-survey.md — finding #4.

What it is

When the agent makes an edit, the system runs the project's linter and lets the agent auto-fix any errors it created. Lint-fix calls are not billed against the user's prompt budget.

Seen in

Windsurf/Cascade ("Linter integration"); Aider has a similar "lint and test" loop.

ocman angle

ocman doesn't bill prompts, but the user friction is real: people get an OpenCode response, then have to manually re-prompt "now fix the lint errors". ocman could expose a per-project "auto-lint hook" toggle: after every assistant turn, run make lint (or a user-configured command) in the project dir; if it fails, send the output back to OpenCode as a follow-up prompt. Reuse the existing tmux launch path.

Effort

S. One hook config + one post-message watcher.

From [spec/multi-agent-support/agent-feature-survey.md](../src/branch/main/spec/multi-agent-support/agent-feature-survey.md) — finding #4. ## What it is When the agent makes an edit, the system runs the project's linter and lets the agent auto-fix any errors it created. Lint-fix calls are *not billed* against the user's prompt budget. ## Seen in Windsurf/Cascade ("Linter integration"); Aider has a similar "lint and test" loop. ## ocman angle ocman doesn't bill prompts, but the *user friction* is real: people get an OpenCode response, then have to manually re-prompt "now fix the lint errors". ocman could expose a per-project "auto-lint hook" toggle: after every assistant turn, run `make lint` (or a user-configured command) in the project dir; if it fails, send the output back to OpenCode as a follow-up prompt. Reuse the existing tmux launch path. ## Effort S. One hook config + one post-message watcher.