#075
Codex
codex exec resume --last enables two-stage pipelines without re-sending context
Chain: codex exec "find the bugs" then codex exec resume --last "fix them". Each step continues the same session with original transcript and approvals intact. Zero re-sending cost.
"If you need to continue a previous run (for example, a two-stage pipeline): codex exec 'review the change' codex exec resume --last 'fix the race conditions you found'"
↗ Source
#076
Codex
codex exec requires a Git repo — running outside one causes a refusal
Running codex exec outside a Git repo causes Codex to refuse execution as a safety guard. Use --skip-git-repo-check only when you're certain the environment is safe.
"Codex requires commands to run inside a Git repository to prevent destructive changes. Override this check with codex exec --skip-git-repo-check."
↗ Source
#077
Codex
The official openai/codex-action GitHub Action handles CI without installing the CLI
Store OPENAI_API_KEY as a GitHub secret, check out your code, and pass prompts inline or via a committed prompt file. The action handles CLI installation — no CLI setup on the runner.
"If you want to avoid installing the CLI yourself, you can run codex exec through the Codex GitHub Action and pass the prompt as an input."
↗ Source
#078
Codex
Auto-fix CI failures: trigger on workflow_run with conclusion == 'failure'
Use on: workflow_run: types: [completed] with the failure condition. Check out the failing SHA, run Codex with a narrow prompt, re-run tests, and open a PR with the patch automatically.
"Trigger a follow-up workflow when your main CI workflow completes with an error. Check out the failing commit SHA. Install dependencies and run Codex with a narrow prompt."
↗ Source
#079
Codex
Treat ~/.codex/auth.json like a password — never commit, share, or log it
Contains access tokens for Codex account auth in CI. Seed through secure storage, run Codex to refresh it in place, and persist the updated file between runs. Don't use this for public repos.
"Treat ~/.codex/auth.json like a password: it contains access tokens. Don't commit it, paste it into tickets, or share it in chat."
↗ Source
#080
Codex
Automations run on worktrees by default — switch to local for live file editing
Scheduled automations use a dedicated background worktree by default, isolating changes from your working copy. Switch to local mode when you want the automation to edit files in your active checkout.
"Use worktrees when you want to isolate automation changes from unfinished local work. Use local mode when you want the automation to work directly in your main checkout."
↗ Source
#081
Codex
High-frequency automations accumulate worktrees — archive old runs to control disk usage
Each automation run creates a fresh copy with its own files, deps, and build caches. Codex keeps 15 by default. For high-frequency schedules, archive completed runs or disk usage grows unbounded.
"frequent schedules can create many worktrees over time. Archive automation runs you no longer need."
↗ Source
#082
Codex
Use $skill-name in automation prompts to explicitly invoke skills
Automation prompts support $skill-name syntax to explicitly trigger a skill. Update the skill once and all automations using it update automatically. No automation config edits needed.
"You can explicitly trigger a skill as part of an automation by using $skill-name inside your automation."
↗ Source
#083
Codex
Automation worktrees run in detached HEAD state — no branches are polluted
Codex creates automation worktrees in detached HEAD state. No branches are created unless you explicitly run 'Create branch here.' Multiple worktrees can exist without polluting your branch namespace.
"The worktree will not be checked out as a branch. It will be in a detached HEAD state. This lets Codex create several worktrees without polluting your branches."
↗ Source
#084
Codex
Setup scripts auto-run when a new worktree is created — define platform-specific variants
Setup scripts run automatically when Codex creates a new worktree. Define platform-specific scripts for macOS, Windows, or Linux. Worktrees need full setup because they're fresh checkouts without node_modules.
"Setup scripts run automatically when Codex creates a new worktree at the start of a new thread."
↗ Source
#085
Codex
Enable Automatic Reviews to review every new PR without a manual @codex comment
Without this setting, Codex only reviews PRs when explicitly tagged. Enable it and Codex posts a review on every newly opened PR. Codex applies AGENTS.md Review Guidelines closest to each changed file.
"If you want Codex to review every pull request automatically, turn on Automatic reviews in Codex settings."
↗ Source
#086
Claude Code
--append-system-prompt-file injects task-specific context in headless/CI mode
Inject system prompt content via this flag without touching CLAUDE.md. Useful for CI pipelines that need task-specific context ('this is a security audit run, apply extra scrutiny') injected at the call site.
"Documentation to clarify that --append-system-prompt-file and --system-prompt [are available for headless mode customization]."
↗ Source
#087
Claude Code
CLAUDE_CODE_USE_FOUNDRY=1 or CLAUDE_CODE_USE_VERTEX=1 routes SDK to cloud
These environment variables alone switch the Agent SDK between Anthropic direct, Azure AI Foundry, and GCP Vertex AI. No code changes, no config file updates.
"Microsoft Azure: set CLAUDE_CODE_USE_FOUNDRY=1 environment variable and configure Azure credentials."
↗ Source
#088
Claude Code
Vertex AI requires both Sonnet AND Haiku quota in us-east5 — missing Haiku causes cryptic errors
Even when you only use Sonnet, Claude Code on Vertex requires access to both Sonnet and Haiku in us-east5 specifically. Haiku handles internal operations like compaction.
"Claude Code on Vertex AI currently only supports the us-east5 region. Users will need access to both Claude 3.7 Sonnet and Claude 3.5 Haiku models in their Vertex AI project."
↗ Source