From zipsa-commands
Implement tasks incrementally with Codex as an approach reviewer. For each task, Agent and Codex independently design the implementation approach, compare, then Agent implements. Use after zipsa-tasks produces a task list. Pass the path to the tasks file or a single task description as the argument.
How this skill is triggered — by the user, by Claude, or both
Slash command
/zipsa-commands:zipsa-buildThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
`$ARGUMENTS` should be the path to the tasks file (e.g. from `zipsa-tasks` output), or a single task description. If a tasks file is provided, tasks are processed one by one in dependency order.
$ARGUMENTS should be the path to the tasks file (e.g. from zipsa-tasks output), or a single task description. If a tasks file is provided, tasks are processed one by one in dependency order.
Repeat Steps 1–4 for each task. Complete each task fully before moving to the next.
Do both at the same time:
1a. Read the task.
Extract from the task:
Identify files likely involved by reading relevant parts of the codebase. Do not implement yet.
1b. Spawn a background subagent for Codex (Task tool with run_in_background: true). The subagent should:
mcp__validate-plans-and-brainstorm-ideas__codex with:
prompt: construct exactly as shown belowsandbox: read-onlyapproval-policy: nevercwd: (use the current working directory)threadId and the full response.First, explore the codebase in your working directory to understand the project structure, key files, and existing patterns.
You are acting as a **Senior Engineer reviewing an implementation approach** — not writing code. For the following task, describe how you would implement it:
- Which files and functions would you modify or create?
- In what order would you make the changes?
- What edge cases or conflicts with existing code should be watched for?
- Is there a simpler alternative approach worth considering?
Do NOT write code. Describe the approach only. Be concise and specific.
Task:
<task>
{paste the full task description and acceptance criteria}
</task>
Skip Codex (Step 1b) when:
While Codex works in the background, independently design your implementation approach:
Do NOT read the Codex response until your own approach is fully formed.
Read the Codex response. Compare it against your own approach:
State the confirmed approach in one short paragraph before implementing.
4a. Implement.
Write the code according to the confirmed approach. Follow project conventions:
4b. Verify acceptance criteria.
Check each acceptance criterion from the task:
4c. On failure:
If verification fails:
mcp__validate-plans-and-brainstorm-ideas__codex-reply to share the failure and ask for a revised approach4d. On success: proceed to the next task.
## Build Complete
### Tasks Implemented
1. [Task title] — ✅ verified
2. [Task title] — ✅ verified
...
### Approach Decisions
- [Task title]: chose [approach] over [Codex alternative] because [reason]
- [Task title]: adopted Codex suggestion to [simplification] — reduced [X] lines
...
### What Was Skipped and Why
- [Task title]: skipped Codex review — [reason]
### Remaining Work
- [anything explicitly out of scope or deferred]
npx claudepluginhub mojokb/zipsa-co-command --plugin zipsa-commandsCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.