thwap/kronai

Watch

Fork 0

feat(engine): implement job lifecycle execution #93

Merged

fuzzy merged 3 commits from feature/0030-implement-job-lifecycle into main

2026-07-05 14:28:36 +00:00

automation commented

2026-07-05 14:18:59 +00:00

Owner

What

Implement the full job lifecycle: read prompt, connect MCP servers, call LLM with tools, log results, and handle ephemeral vs recurring cleanup.

Why

Completes the Phase 2 job execution engine — kronai can now run jobs end-to-end.

Testing

78 unit tests pass (8 new for job lifecycle)
Pre-commit hooks pass (ruff + mypy strict)
Coverage: 88%

Breaking Changes

None.

Notes

Ephemeral jobs are deleted after successful execution. Recurring jobs persist. Results written to {job_path}/results/{timestamp}/output.md.

Closes #0027
Closes #0028
Closes #0029
Closes #0030

## What Implement the full job lifecycle: read prompt, connect MCP servers, call LLM with tools, log results, and handle ephemeral vs recurring cleanup. ## Why Completes the Phase 2 job execution engine — kronai can now run jobs end-to-end. ## Testing - [x] 78 unit tests pass (8 new for job lifecycle) - [x] Pre-commit hooks pass (ruff + mypy strict) - [x] Coverage: 88% ## Breaking Changes None. ## Notes Ephemeral jobs are deleted after successful execution. Recurring jobs persist. Results written to `{job_path}/results/{timestamp}/output.md`. Closes #0027 Closes #0028 Closes #0029 Closes #0030

automation added 3 commits

2026-07-05 14:18:59 +00:00

feat(engine): implement job lifecycle execution 4ea54befff

- run_job: read prompt, connect MCP servers, call LLM, log results
- Ephemeral jobs: delete job directory after successful run
- Recurring jobs: keep directory after run
- _write_result: timestamped output files in results/ subdirectory
- run_interval: discover and execute all jobs for a given interval
Closes #0027
Closes #0028
Closes #0029

test(engine): add unit tests for job lifecycle c5531ada35

Cover recurring jobs, ephemeral deletion, missing prompts,
LLM errors, MCP errors, system prompt inclusion, interval
execution, and empty intervals.

docs: mark Phase 2 Task 4 complete

CI / test (pull_request) Successful in 50s

Details

CI / lint (pull_request) Successful in 1m20s

Details

d51526848e

Job lifecycle with ephemeral/recurring execution, result logging,
and directory cleanup done.
Closes #0030

the.auditor approved these changes

2026-07-05 14:26:22 +00:00

the.auditor left a comment

Automated Review — PR #93

Result: APPROVED

The capstone of Phase 2 — wires config, discovery, prompt reader, LLM client, and MCP integration into a working job lifecycle. Clean orchestration.

Correctness: Full lifecycle works end-to-end. System prompt prepended to user prompt. MCP tools converted and passed to LLM. Ephemeral jobs deleted on success. Results written to timestamped dirs.
Completeness: Closes #27–#30. ROADMAP updated.
Test coverage: 8 tests covering success (recurring + ephemeral), missing prompt, LLM error, MCP error, system prompt inclusion, interval dispatch, and empty intervals.
Error handling: JobError, LLMError, MCPError all caught per-job with logging and graceful return (False). MCP connections cleaned up on error paths.
Maintainability: Good function separation. _write_result and _cleanup_mcps extracted as helpers. Clear flow in run_job().

Suggestions (non-blocking)

src/kronai/engine.py:68,133 — load_config() is called in both run_job() and run_interval(). For an interval with N jobs, config is loaded N+1 times. Consider loading once in run_interval() and passing the config to run_job().
src/kronai/engine.py:98-104 — When the LLM responds with tool calls (function/tool_use), _extract_content in llm.py would return "None" (since content is null). The engine should eventually handle tool call execution rather than writing "None" as the result.
src/kronai/engine.py:113 — shutil.rmtree(..., ignore_errors=True) could fail silently. Consider adding a logged warning or checking success:
```
shutil.rmtree(job.path)
logger.info("Job %d ephemeral directory deleted", job.id)
```
src/kronai/engine.py:166 — server._name accessed via private attribute. Consider adding a name property to MCPServer or using server._name with a public accessor. (Already suppressed with noqa: SLF001.)
src/kronai/engine.py:140 — Timestamp in _write_result uses local time via time.strftime. Consider using UTC (time.gmtime) for reproducible timestamps across environments.

## Automated Review — PR #93 **Result: APPROVED** --- The capstone of Phase 2 — wires config, discovery, prompt reader, LLM client, and MCP integration into a working job lifecycle. Clean orchestration. - **Correctness**: Full lifecycle works end-to-end. System prompt prepended to user prompt. MCP tools converted and passed to LLM. Ephemeral jobs deleted on success. Results written to timestamped dirs. - **Completeness**: Closes #27–#30. ROADMAP updated. - **Test coverage**: 8 tests covering success (recurring + ephemeral), missing prompt, LLM error, MCP error, system prompt inclusion, interval dispatch, and empty intervals. - **Error handling**: `JobError`, `LLMError`, `MCPError` all caught per-job with logging and graceful return (False). MCP connections cleaned up on error paths. - **Maintainability**: Good function separation. `_write_result` and `_cleanup_mcps` extracted as helpers. Clear flow in `run_job()`. ### Suggestions (non-blocking) 1. **`src/kronai/engine.py:68,133`** — `load_config()` is called in both `run_job()` and `run_interval()`. For an interval with N jobs, config is loaded N+1 times. Consider loading once in `run_interval()` and passing the config to `run_job()`. 2. **`src/kronai/engine.py:98-104`** — When the LLM responds with tool calls (function/tool_use), `_extract_content` in `llm.py` would return `"None"` (since `content` is null). The engine should eventually handle tool call execution rather than writing `"None"` as the result. 3. **`src/kronai/engine.py:113`** — `shutil.rmtree(..., ignore_errors=True)` could fail silently. Consider adding a logged warning or checking success: ```python shutil.rmtree(job.path) logger.info("Job %d ephemeral directory deleted", job.id) ``` 4. **`src/kronai/engine.py:166`** — `server._name` accessed via private attribute. Consider adding a `name` property to `MCPServer` or using `server._name` with a public accessor. (Already suppressed with `noqa: SLF001`.) 5. **`src/kronai/engine.py:140`** — Timestamp in `_write_result` uses local time via `time.strftime`. Consider using UTC (`time.gmtime`) for reproducible timestamps across environments.

the.auditor referenced this pull request

2026-07-05 14:26:26 +00:00

engine: load_config called N+1 times in run_interval #94

the.auditor referenced this pull request

2026-07-05 14:26:27 +00:00

engine: handle LLM tool call responses #95

the.auditor referenced this pull request

2026-07-05 14:26:29 +00:00

engine: ephemeral deletion may fail silently #96

the.auditor referenced this pull request

2026-07-05 14:26:30 +00:00

mcp: add public name property to MCPServer #97