// Run the full development pipeline - groom, implement, test, accept, commit, repeat
| name | execute |
| description | Run the full development pipeline - groom, implement, test, accept, commit, repeat |
| disable-model-invocation | true |
Run the full development pipeline for codehive.
Argument: [number-of-issues] — how many issues per batch (default: 2)
Before starting, read docs/PROCESS.md for the full workflow.
Find all .todo.md files in docs/tracker/. For each, launch the product-manager agent in grooming mode. This can run in background while other work proceeds.
List .groomed.md files in docs/tracker/. Pick the lowest-numbered $1 issues (default 2). Check dependencies — skip if deps aren't in docs/tracker/done/. Skip issues with [BLOCKED] in title.
For each picked issue, create task panel items with dependencies:
[PM groom #N] -> [SWE #N] -> [QA #N] -> [PM accept #N] --\
+--> [Pull next]
[PM groom #M] -> [SWE #M] -> [QA #M] -> [PM accept #M] --/
Set blockedBy dependencies between tasks. Mark in_progress when starting, completed when done.
For each issue, launch software-engineer agent with the .groomed.md filename. One agent per issue — never combine.
After engineer reports done, launch tester agent with the .in-progress.md filename. One agent per issue.
Launch product-manager agent in acceptance mode for each passing issue. One agent per issue.
For each accepted issue:
mv docs/tracker/NN-name.in-progress.md docs/tracker/done/NN-name.done.mdgit add .git commit -m "Implement issue NN: short description"After pushing, launch oncall-engineer to verify CI/CD is green.
Go back to step 2. Pick next batch. Continue until no more .groomed.md or .todo.md issues remain.
IMPORTANT: The orchestrator NEVER writes code. It only launches agents, routes feedback, manages task panel items, and commits after PM accepts.