← Back to Daily Builds

Daily build note · May 22, 2026

Codex Skill Ops Sprint

A fixed-scope service and starter kit that turns ad hoc coding-agent usage into reusable Codex skills, automations, MCP/tool policies, review gates, and cost-aware runbooks for small...

Mixed Wildcard - Combine Or Alternate Between The Strongest Themes Above Build note published Public demo coming soon

What shipped

Built a runnable local Python CLI kit named codex-skill-ops-sprint. It can load a sample or custom workspace, recommend Codex skills, and generate a staged repo package with:

  • Three default SKILL.md files under .codex/skills/.
  • AGENT_OPS.md.
  • agent-ops-register.json.
  • mcp-tool-policy.md.
  • pr-review-gates.md.
  • cost-hygiene.md.
  • sprint-summary.md.

The smoke path uses the Acme SaaS sample and verifies expected files, required sections, and policy defaults for write-capable tools.

Architecture

  • Python standard library only, so the demo runs without dependency installation or network access.
  • Dataclasses plus JSON files for transparent local persistence.
  • Deterministic generation in skillops/generator.py instead of live LLM calls.
  • Starter skill definitions in skillops/library.py, separate from workspace persistence and CLI wiring.
  • Exports are staged under exports/<workspace-id>/repo-package/ and require --force to replace.

Trimmed scope

  • No web UI.
  • No live GitHub, MCP, browser, billing, or provider integrations.
  • No PDF or zip export.
  • No automatic install into a client repo.
  • No external AI model calls or secrets.

Limitations

  • Custom workspaces created with init are minimal; richer intake is easiest by editing JSON or using the sample as a template.
  • Markdown rendering is deterministic but not template-file driven.
  • Policy guidance is practical operating guidance, not security, legal, or compliance certification.
  • Recommendations are profile-based with light workflow enrichment, not ML-based.

Verification

Run:

``bash python -m skillops smoke ``

Expected result: the command prints Smoke passed and writes the sample package to exports/acme-saas/repo-package/.

Suggested next steps

  • Add an interactive intake command for editing workflows, tool policies, and automations without hand-editing JSON.
  • Add optional importers for PR templates, CI commands, and issue labels.
  • Add checked examples for frontend agency and technical consulting automation workspaces.
  • Add a simple preview UI after the CLI package proves useful in real sprint delivery.
  • Add zip export for client handoff.