codex-implementation
Implements scoped code changes with Codex CLI, then inspects the diff.
Skill by connorch
Chalk Labs
Need help deploying codex-implementation?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/connorch-codex-implementation
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: productivity, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Nov 1, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/codex-implementation
What's inside
- Skill name
codex-implementation- Description
- Implements scoped code changes with Codex CLI, then inspects the diff. This is how the GPT model rated intelligence 9 in the model-selection table is invoked for implementation work. Use when the user asks to delegate implementation to Codex or a GPT model, when the model-selection rubric routes the work to that model, or when a bounded task would benefit from another coding agent producing a patch.
- License
- MIT
- Instructions
- 96 lines in SKILL.md · view on GitHub
Files (2)
- SKILL.md
- openai.yaml · agents/
SKILL.md (first lines)
# Codex Implementation
Use Codex as a separate implementation agent for bounded code changes. Claude remains responsible for scoping the task, reviewing the diff, running or checking verification, and explaining the final result.
Use this when the user asks for Codex or delegation, or when a bounded task would benefit from a parallel implementation agent producing a patch. Do not let Codex commit, push, deploy, or edit global config unless the user explicitly asked for that.
## Workflow
1. Pin the current state with `git status --short` and note any user changes already present.
2. Define the implementation scope: files or behavior to change, files to avoid, constraints, and verification commands.
3. Create a temporary artifact directory for Codex's report.
4. Run `codex exec` with repo write access.
5. Confirm `run.log` has content shortly after launch.
6. After Codex exits, inspect `git status` and `git diff`.
7. Run the cheapest reliable verification yourself when practical.
8. Report what Codex changed, what Claude verified, and any remaining risks.
Use this command shape:
```bash
ARTIFACT_DIR="$(mktemp -d "${TMPDIR:-/tmp}/codex-implementation.XXXXXX")"
REPORT="$ARTIFACT_DIR/report.md"
PROMPT="$ARTIFACT_DIR/prompt.md"
# Write a self-contained prompt to $PROMPT, then run:
codex exec \
-C "$PWD" \
--add-dir "$ARTIFACT_DIR" \
-s workspace-write \
-o "$REPORT" \
"$(cat "$PROMPT")" </dev/null >"$ARTIFACT_DIR/run.log" 2>&1
```
Use `-s workspace-write` by default. Use `-s danger-full-access` only when the implementation truly needs access outside the repo, app launch automation, simulator work, package-manager global state, or other machine-level operations.
If `codex` is not installed or the command fails, report the error and offer to implement the change directly instead.
### Running in the background
Runs commonly exceed the Bash tool's 10-minute timeout. Either pass an explicit long timeout or run in the background.
- Always close stdin with `</dev/null`, backgrounded or not. Codex otherwise blocks waiting on a terminal that never answers and the run hangs silently.
- Shortly after launch, check once that `run.log` has content. This catches a bad launch (missing binary, rejected flag) without waiting for the full run. After that, rely on the process exit notification; no polling.
- The report file at `$REPORT` is the completion signal. Do not read the diff until it exists.
## Prompt Requirements
Tell Codex:
- The exact task and expected behavior.
- The repo path and current branch context if relevant.
- That it must preserve unrelated user changes.
- That it must not commit, push, deploy, or edit global config.
- Which verification commands to run, or to explain why they were skipped.
- To write a concise final report with files changed, verification, and unresolved questions.
Keep the task bounded. If the requested work bundles several substantial changes, split it into separate Codex runs or ask the user to choose the first scope.
## Example Prompt
```text
You are implementing a scoped change for Claude.
Repository: /absolute/path/to/repo
Artifact directory: /tmp/codex-implementation.XXXXXX
Goal:
- Add keyboard navigation to the command palette.
Acceptance criteria:
- ArrowUp and ArrowDown move the highlighted item.
- Enter selects the highlighted item.
- Escape closes the palette.
- Existing mouse behavior keeps working.
Constraints:
- Preserve unrelated user changes.
- Do not commit, push, deploy, or edit global config.
- Follow existing component and test patterns.
Verification:
- Run the focused component tests if available.
- Otherwise run the nearest relevant typecheck or test command and explain the choice.
Report:
- Files changed
- Behavioral summary
- Verification run and result
- Anything blocked or uncertain
```
## Review After Codex
Always inspect Codex's diff before telling the user the work is done. Revert only Codex-created mistakes when you are sure they are not user changes. If Codex leaves the repo in a worse state or changes unrelated files, stop and report the issue with the diff summary.From connorch/skills, MIT.
More from connorch/skills
- 0.603score
qa-ux-verify
connorch
Runs a QA plan in a browser and reports the results without fixing anything.
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
step-back
connorch
Step back from a branch that grew through iteration - review rounds, point fixes, PR feedback - and judge the end state as a whole: are the big architecture…
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
codex-review
connorch
Gets an independent code review from Codex CLI.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
fleet-exec
connorch
Run commands on another of the user's machines (the macOS and linux machines on their tailnet) over Tailscale SSH.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
implement-and-review
connorch
Use when the user approves a plan and says to implement it.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
claude-router
connorch
Controls claude-router, the local proxy that spreads Claude Code requests across subscription accounts.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
mac-vm
connorch
Use when a task needs a GUI app on a Mac whose desktop is locked, or when the user asks to start, stop, or set up the agent macOS VM.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
merge-risk
connorch
Rates how risky it is to merge a pull request or branch, on a Minimal / Low / Moderate / High / Critical scale, with the blockers that set the rating.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
pull-upstream
connorch
Sync a forked repository with its upstream.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
qa-ux-plan
connorch
Writes an end-to-end QA plan from the branch diff without running tests.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
t3-browser
connorch
Use for any browser QA, verification, screenshot, or "check it in the browser" step when the T3 Code Browser panel tools (mcp__t3-code__preview_*, such as…
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
claude-code-subagent
connorch
Delegate a bounded task to Claude Code CLI (`claude -p`) as a headless subagent, then review what it produced.
SkillAFreshness: A★ 2Claude Code +9
Signals
- Stars
- 2
- Forks
- 0
- Open issues
- 2
- Last commit (this folder)
- Oct 5, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/connorch/skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from connorch
- 0.603score
step-back
connorch
Step back from a branch that grew through iteration - review rounds, point fixes, PR feedback - and judge the end state as a whole: are the big architecture…
SkillAFreshness: A★ 2Claude Code +9 - 0.603score
qa-ux-verify
connorch
Runs a QA plan in a browser and reports the results without fixing anything.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
claude-router
connorch
Controls claude-router, the local proxy that spreads Claude Code requests across subscription accounts.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
html-communication
connorch
Writes HTML writeups and publishes them privately to files.vaut.dev.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
fleet-exec
connorch
Run commands on another of the user's machines (the macOS and linux machines on their tailnet) over Tailscale SSH.
SkillAFreshness: A★ 2Claude Code +9 - 0.598score
codex-review
connorch
Gets an independent code review from Codex CLI.
SkillAFreshness: A★ 2Claude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get codex-implementation running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
codex-implementation skill. Agent Mart. https://agentmart.dev/skills/connorch-codex-implementation (last updated 2026-10-24)