| name | zeus-execute |
| description | Execute all pending tasks in task.json using the zeus-runner.sh shell loop. Tasks are grouped into dependency waves and run with fresh claude CLI subagents. Each task gets an atomic git commit and a mandatory ai-log entry. Supports --version flag for v2/v3/v4 execution. |
Zeus Execute - Wave Orchestration Engine
Run pending tasks with dependency awareness, strict quality gates, and atomic commits.
Preconditions
.zeus/{version}/config.json exists.
.zeus/{version}/task.json exists and contains pending tasks unless filtered by --task.
.zeus/scripts/zeus-runner.sh exists and is executable.
claude CLI is installed and reachable.
Optional filters
--task T-###
--wave <N>
--version vN
Deterministic workflow
1) Build execution plan
Load pending tasks and group by wave.
Show preview summary before running:
- total task count
- wave count
- per-wave task list
Require user confirmation to proceed.
2) Launch runner
bash .zeus/scripts/zeus-runner.sh \
--task-file .zeus/{version}/task.json \
--log-dir .zeus/{version}/ai-logs \
--version {version}
3) Monitor and report
Track runner output and report progress by wave:
- started
- completed
- failed/retry state
- commit SHA for completed tasks
4) Validate completion
After run, verify:
- all targeted tasks now have
passes: true,
commit_sha is filled,
- milestone status can be promoted to
completed when all related tasks pass.
5) Persist post-run updates
Update relevant roadmap milestone status and write summary log.
6) Write execution AI log
Create .zeus/{version}/ai-logs/{ISO-ts}-execute.md.
Per-task execution contract
Each task completion must include:
- Quality checks (typecheck/lint/tests when configured)
- Atomic commit with format
{type}({task_id}): {description}
task.json update (passes, commit_sha, optional ai_log_ref)
- Task-level AI log
.zeus/{version}/ai-logs/{ISO-ts}-{task_id}.md
- Progress append to
.zeus/{version}/progress.txt
Brownfield extension:
- If task includes
refactor_of, switch to minimal-intrusion strategy:
- Add or update tests around existing behavior first.
- Apply smallest safe change in existing modules.
- Preserve backward compatibility or provide migration notes.
- Include
refactor_of details in task-level AI log execution summary.
Failure policy
- quality check failure: pause and surface error details,
- repeated failure (
>3 retries): stop current wave and request task split,
- interruption: resume by skipping already passed tasks.
Output contract
- execution summary by wave,
- list of completed tasks + commit SHAs,
- recommended next action (
/zeus:feedback or /zeus:status).
Optional post-wave test refresh
After each wave completes, offer:
Wave N complete. Refresh test flows for completed tasks? [yes/no]
(Runs: bash .zeus/scripts/generate-tests.sh --version {version} --force)
This regenerates {platform}.test.json scenarios for tasks marked passes: true in the current wave, keeping test coverage aligned with shipped code.
Agent collaboration
Use zeus-executor to supervise multi-wave execution status and enforce completion gates.
Use zeus-tester (via /zeus:test-gen) after execution waves to generate or refresh platform test flows.