Skip to content

v0.4.0 — Codex CLI support

Choose a tag to compare

@iamlaforge iamlaforge released this 06 May 00:47
· 6 commits to main since this release

The third harness: Spruce now works in Codex CLI in addition to Claude Code and Cursor. Codex picks up Spruce skills natively from the project's .agents/skills/ directory — Codex's primary skill-loading path — and follows each skill's
conversation pattern correctly. Verified end-to-end via real Codex CLI smoke tests.

Same install command:

npx spruce-skill add

Auto-detects which AI coding tool the project uses and installs the right per-harness bundle. For Codex users, files land at .agents/skills/<command>/SKILL.md — exactly where Codex CLI reads from.

No code changes were required.

The v0.3.0 architecture committed to a thin install wrapper around vercel-labs/skills plus shipping standard Anthropic Agent Skills format to each harness's documented path. Codex working out of
the box confirms the architecture scales — each new harness compatible with the Agent Skills spec should "just work" via the same install command.

Invocation patterns in Codex CLI (all verified working):

  • Multiple ways. Type $decide, $typeface, $personas, etc. to launch a specific skill directly. Or type /to browse the skill picker tree. Or describe intent in natural language — Codex’s matcher selects the relevant skill from descriptions.-

Same 25-command catalog, same conversation patterns, same .spruce.md context grounding — calibrated to whatever project Codex is opened in.

Smoke test confirmation:

In a real Codex CLI session, asking "Fix the typography on a pricing page for this product" led Codex to:

  1. Recognize the typography intent and explicitly invoke the typeface skill
  2. Read SKILL.md from .agents/skills/typeface/
  3. Read the project's .spruce.md for context
  4. Apply Spruce-calibrated typography reasoning to the work

The /decide skill similarly drove Codex through the one-decision-at-a-time interview pattern, asking only the first decision and waiting — exactly per the skill's instructions.

Preserved from v0.3.0:

  • All 25 commands across 5 tiers (Setup, Discovery, Generative, Diagnostic, Corrective) work identically.
  • Claude Code slash commands at .claude/commands/ still ship alongside the skills installation; the existing slash UX is unchanged.
  • The 13 reference files (Discovery foundation + design dimensions) bundle into each skill so context-rich reasoning travels with every command.
  • The .spruce.md context file remains the foundation every command reads from.