Run Claude Code CLI in interactive or headless mode for coding, editing, review, discussing or any task delegation
Use this skill to run Claude Code CLI for coding, editing, review, discussion, or task delegation.
Last verified against claude --help on 2026-02-20.
claude, claude -p, or similar direct CLI entrypoints for agent work.status, result, and resume against the tracked job id.--wait as compatibility-only; it polls tracked state and is not the primary review flow.For this repository, the allowed agent path is the tracked wrapper:
bash .agents/skills/claude/scripts/run-claude-runtime.sh ...
Use the wrapper instead of direct claude shell invocations when operating in this repo.
Primary path for this repo:
bash .agents/skills/claude/scripts/run-claude-runtime.sh start \
--cwd /path/to/worktree \
--prompt-file /tmp/review-prompt.txt \
--json
Follow-up control:
bash .agents/skills/claude/scripts/run-claude-runtime.sh status \
--cwd /path/to/worktree \
--job-id <job-id> \
--json
bash .agents/skills/claude/scripts/run-claude-runtime.sh result \
--cwd /path/to/worktree \
--job-id <job-id> \
--json
Optional follow-up if you need to continue a prior tracked run:
bash .agents/skills/claude/scripts/run-claude-runtime.sh resume \
--cwd /path/to/worktree \
--job-id <job-id> \
--json
Compatibility path only:
bash .agents/skills/claude/scripts/run-claude-runtime.sh start \
--cwd /path/to/worktree \
--prompt-file /tmp/review-prompt.txt \
--wait
Rules:
start, then status and result, over any waiting path--wait as a bounded convenience layer over tracked job state, not as proof of Claude livenesstimed out waiting for completion, not as a Claude hang--resume-last; the tracked runtime intentionally keeps resume resolution deterministic| Use case | Command | Notes |
|---|---|---|
| Tracked review start | bash .agents/skills/claude/scripts/run-claude-runtime.sh start ... | Preferred repo-local review path. Returns a tracked job id unless --wait is used. |
| Tracked review status | bash .agents/skills/claude/scripts/run-claude-runtime.sh status ... | Reads persisted job state instead of inferring progress from stdout timing. |
| Tracked review result | bash .agents/skills/claude/scripts/run-claude-runtime.sh result ... | Fetches final Claude output only after the job reaches a terminal state. |
| Tracked review resume | bash .agents/skills/claude/scripts/run-claude-runtime.sh resume ... | Resume by tracked job id or explicit session id. --resume-last stays unsupported on purpose. |
| Compatibility wait | bash .agents/skills/claude/scripts/run-claude-runtime.sh start ... --wait | Secondary path only. Polls tracked state and may report timed out waiting for completion. |