Automation: JSON output, a Python API and a GitHub Action for self-hosted runners #48

Activity

1. nasser1941

   added

   enhancementNew feature or requestNew feature or request

   on Oct 1, 2026
2. nasser1941

   closed this as completedin #81on Oct 5, 2026
3. nasser1941 commented on Oct 5, 2026

   

   nasser1941

   on Oct 5, 2026

   OwnerAuthor

   More actions

   Shipped in 0.18.0 (#81). See Automation.

   • lcode -p "…" --output json | stream-json, with --max-steps and --allowed-tools.
   • from lcode import Session, with an example in examples/api/fix_tests.py.
   • uses: nasser1941/lcode@v0.18.0 for self-hosted runners: it answers @lcode and can review pull requests.

   Tested with Qwen3.6 35B through Ollama, also from the PyPI build:

   • --output json: fixed the bugs and reported calc.py changed, with the usage.
   • --output stream-json --max-steps 3: the expected events, and exit code 3.
   • The Python example: it approved only the test command; done in 5 model requests.
   • lcode action on a simulated @lcode issue comment, against a stand-in GitHub API and a bare remote.
     • lcode reacted to the comment and ran the tests (allow-commands), then pushed a branch with the fix as lcode.
     • It opened a pull request that closes the issue and replied with a summary.
     • Earlier versions had the model try to commit and then say so. Now git is denied to the model and the reply is written after the change is proposed.

   Not yet tried on a real GitHub self-hosted runner: that needs a runner registered to a repository. Reports from anyone who sets it up are welcome here.