Claude Code Agent Skill

Operate Claude Code as a managed coding agent — from worktree setup through PR merge.

Prerequisites

claude --version  # Verify installed
# Install: npm install -g @anthropic-ai/claude-code
tmux -V             # tmux required for full workflow

CLI Quick Reference

Execution Modes

Quick Mode — Small Tasks

For trivial fixes, one-file changes, or analysis. Use -p (non-interactive).

Output capture: Always redirect output to a log file so it's readable regardless of PTY availability. Use --output-format stream-json for structured, parseable progress events (each message arrives as a separate JSON line).

LOG_FILE="/tmp/claude-quick-${TASK_ID:-$$}.log"

# Via OpenClaw exec — use background=true + pty=true, NO hard timeout
# pty=true ensures claude CLI flushes output properly (no buffering issues)
# (hard timeout kills the process; instead we poll and extend)
# Redirect both stdout and stderr to log file via tee so output is always captured.
# In -p mode (non-interactive), | tee is safe — no TTY detection issues.
exec(command="claude -p 'fix the typo in README.md' --dangerously-skip-permissions --output-format stream-json 2>&1 | tee -a $LOG_FILE",
     workdir="/path/to/project", background=true, pty=true)

PTY fallback: If pty=true is unavailable (some containers, CI runners), the command still works because -p mode is non-interactive — it doesn't rely on isatty(stdout). The 2>&1 | tee ensures both stdout and stderr are captured to the log file regardless of PTY status. Without PTY, you lose color output but all content is preserved.

Adaptive Timeout (Poll-and-Extend)

Do NOT use timeout= for claude tasks. Instead, use background execution with periodic polling. This prevents premature kills on long-running tasks:

1. Launch with background=true (no timeout)
2. Poll every ~5 min with process(action="poll", sessionId=<id>, timeout=300000)
3. If process is still running → check log file for new output
4. If process exited → check exit code and log file, done
5. Safety net: if no new output for 12 hours, ask user before killing
6. Safety net: if output is repeating (loop detection), ask user

Persistent polling state: Store polling metadata in the task registry so a restarted orchestrator agent can resume monitoring without losing state:

Registry fields for Quick Mode tasks:
  "lastOutputHash": "<sha256 of last 20 lines>",
  "lastCheckedAt": <unix timestamp>,
  "silentRounds": <int>,
  "repeatingRounds": <int>

Poll loop (agent behavior, not a script):

  poll_interval = 5 min (300000 ms)
  max_silent_rounds = 144  (= 12 hours with no new output → ask user)
  max_repeating_rounds = 12 (= 1 hour of identical output → likely stuck)

  # Restore state from registry if resuming after agent restart
  silent_rounds = registry[task_id].silentRounds ?? 0
  repeating_rounds = registry[task_id].repeatingRounds ?? 0
  last_output_hash = registry[task_id].lastOutputHash ?? ""

  repeat:
    result = process(action="poll", sessionId=<id>, timeout=300000)
    if result.completed:
      → check exit code, read $LOG_FILE, report result
      → break
    else:
      # Read latest output directly from the log file
      new_output = tail -20 "$LOG_FILE"
      new_hash = sha256(new_output)

      if new_hash != last_output_hash and new_output != "":
        if last_output_hash != "" and output_looks_similar(new_output, last_output):
          repeating_rounds += 1   # output changing but repetitive (loop)
          silent_rounds = 0
        else:
          silent_rounds = 0       # genuinely new output, keep going
          repeating_rounds = 0
        last_output_hash = new_hash
      else:
        silent_rounds += 1

      # Persist state to registry (survives agent restart)
      update_registry(task_id, {
        lastOutputHash: new_hash,
        lastCheckedAt: now(),
        silentRounds: silent_rounds,
        repeatingRounds: repeating_rounds
      })

      if silent_rounds >= max_silent_rounds:
        → notify user: "Claude has been silent for 12 hours, kill or keep waiting?"
        → wait for user decision
      if repeating_rounds >= max_repeating_rounds:
        → notify user: "Claude appears stuck in a loop (1h of repeated output), kill or keep waiting?"
        → wait for user decision

This way tasks that need 5 min or several hours both work without premature kills.

Full Mode — Features, Bugfixes, Refactors

For non-trivial tasks, use the full workflow below. This gives you:

• Isolated worktree — no conflicts with other work
• tmux session — mid-task steering without killing the agent
• Task tracking — know what's running at all times
• Quality gates — Definition of Done checklist
• Smart retries — don't waste tokens on repeated failures

Full Workflow: Task → Merged PR

Step 1: Create Worktree

Isolate each task in its own worktree and branch:

TASK_ID="feat-custom-templates"
BRANCH="feat/$TASK_ID"
REPO_ROOT=$(git rev-parse --show-toplevel)
WORKTREE="/tmp/worktrees/$TASK_ID"

git worktree add -b "$BRANCH" "$WORKTREE" origin/main
cd "$WORKTREE"

# Install dependencies (adapt to your stack)
pnpm install   # or: npm install / go mod tidy / pip install -r requirements.txt

Step 2: Launch Agent in tmux

Start Claude Code in interactive mode (no -p) so you can steer mid-task. Important: Use tmux pipe-pane to log output — do NOT use | tee because it turns stdout into a pipe, which breaks interactive mode (claude detects !isatty(stdout) and may disable interactive features, breaking send-keys steering).

Critical: Set up pipe-pane BEFORE sending the command. Otherwise early output (startup messages, fast crashes) is lost.

LOG_FILE="/tmp/worktrees/$TASK_ID/claude-output.log"
MAX_LOG_SIZE=$((100 * 1024 * 1024))  # 100 MB safety cap

# 1. Create session with an idle shell first
tmux new-session -d -s "$TASK_ID" -c "$WORKTREE"

# 2. Start pipe-pane BEFORE the command runs — captures ALL output from the start
#    Strip ANSI escape codes so log files are clean and grep-parseable
tmux pipe-pane -t "$TASK_ID" -o "sed 's/\x1b\[[0-9;]*[a-zA-Z]//g' >> $LOG_FILE"

# 3. NOW send the command — all output is captured
tmux send-keys -t "$TASK_ID" "claude --dangerously-skip-permissions \
  'Your detailed prompt here.

When completely finished:
1. Commit all changes with descriptive messages
2. Push the branch: git push -u origin $BRANCH
3. Create PR: gh pr create --fill
4. Notify: openclaw system event --text \"Done: $TASK_ID\" --mode now'" Enter

Log file management: For very long-running tasks, the log file can grow large. Monitor its size and rotate if needed:

LOG_SIZE=$(stat -c%s "$LOG_FILE" 2>/dev/null || echo 0)
if [ "$LOG_SIZE" -gt "$MAX_LOG_SIZE" ]; then
  mv "$LOG_FILE" "${LOG_FILE}.old"
  # pipe-pane will create a new file on next write
fi

Why interactive mode (no -p)?

• Allows mid-task steering via tmux send-keys
• Agent can be redirected without killing and restarting
• --dangerously-skip-permissions is safe in container/sandbox environments

Note on stdout/stderr: In tmux, both stdout and stderr from the process flow through the PTY and are captured by pipe-pane. They are mixed together — you cannot separate them after capture. For error diagnosis, grep for keywords like error, fail, panic in the clean (ANSI-stripped) log file.

Step 3: Register Task

Track all active tasks in a JSON registry. Use flock for atomic updates to prevent race conditions when multiple agents run in parallel.

mkdir -p "$REPO_ROOT/.clawd"
TASKS_FILE="$REPO_ROOT/.clawd/active-tasks.json"

# Initialize if not exists
[ -f "$TASKS_FILE" ] || echo '{"tasks":[]}' > "$TASKS_FILE"

# Get the PID of the claude process inside tmux for reliable status checks
PANE_PID=$(tmux display-message -t "$TASK_ID" -p '#{pane_pid}')

# Register — use flock to prevent concurrent write races
(
  flock -x 200
  jq --arg id "$TASK_ID" --arg branch "$BRANCH" --arg wt "$WORKTREE" \
    --arg pane_pid "$PANE_PID" \
    '.tasks += [{
      "id": $id,
      "agent": "claude",
      "branch": $branch,
      "worktree": $wt,
      "tmuxSession": $id,
      "panePid": ($pane_pid | tonumber),
      "status": "running",
      "startedAt": (now|floor),
      "pr": null,
      "retries": 0,
      "checks": {},
      "lastOutputHash": "",
      "lastCheckedAt": (now|floor),
      "silentRounds": 0,
      "repeatingRounds": 0
    }]' "$TASKS_FILE" > /tmp/tasks.$$.json && mv /tmp/tasks.$$.json "$TASKS_FILE"
) 200>"$TASKS_FILE.lock"

Step 4: Monitor & Steer

# --- Process status check (reliable — checks actual process, not just tmux session) ---

# Method 1: Check if the claude process inside the pane is alive
PANE_PID=$(tmux display-message -t "$TASK_ID" -p '#{pane_pid}' 2>/dev/null)
if [ -z "$PANE_PID" ]; then
  echo "tmux session gone"
elif pgrep -P "$PANE_PID" > /dev/null 2>&1; then
  echo "running"
else
  echo "process exited (tmux session still open)"
  # Get exit code from the shell inside tmux
  tmux send-keys -t "$TASK_ID" 'echo "EXIT_CODE=$?"' Enter
fi

# Method 2: Use tmux's pane_dead flag (if remain-on-exit is set)
# tmux display-message -t "$TASK_ID" -p '#{pane_dead}'  # 1 = process exited

# --- View output ---

# Full output history from log file (ANSI-stripped, grep-friendly)
tail -100 "/tmp/worktrees/$TASK_ID/claude-output.log"

# Search for errors in clean log
grep -i "error\|fail\|panic" "/tmp/worktrees/$TASK_ID/claude-output.log"

# Live view (raw tmux pane, may contain ANSI codes — use for quick glance only)
tmux capture-pane -t "$TASK_ID" -p -S -50

# --- Mid-task steering (DON'T kill — redirect!) ---

# Agent going the wrong direction?
tmux send-keys -t "$TASK_ID" "Stop. Focus on the API layer first, not the UI." Enter

# Agent missing context?
tmux send-keys -t "$TASK_ID" "The schema is in src/types/template.ts. Use that." Enter

# Agent's context window filling up?
tmux send-keys -t "$TASK_ID" "Focus only on these 3 files: api.ts, handler.ts, types.ts" Enter

# Agent needs test guidance?
tmux send-keys -t "$TASK_ID" "Run 'npm test -- --grep auth' to verify your changes." Enter

Monitoring cadence: Check every 5-10 minutes, not every 30 seconds. Agents need time to work.

Step 5: Definition of Done

A PR is NOT ready for review until all checks pass:

✅ PR created              → gh pr list --head "$BRANCH"
✅ No merge conflicts       → gh pr view $PR_NUM --json mergeable -q '.mergeable'
✅ CI passing               → gh pr checks $PR_NUM
✅ AI code review passed    → at least one cross-model review (see Step 6)
✅ UI screenshots included  → (if applicable) screenshot in PR description

Quick inline check:

PR_NUM=$(gh pr list --head "$BRANCH" --json number -q '.[0].number')
echo "PR: #$PR_NUM"
gh pr checks "$PR_NUM"
gh pr view "$PR_NUM" --json mergeable -q '.mergeable'

Step 6: Multi-Model Code Review

Review with a different model than the one that wrote the code. Different models catch different issues:

DIFF=$(gh pr diff "$PR_NUM")

# Option A: Codex reviews Claude's code (best for edge cases & logic errors)
echo "$DIFF" | codex exec -s read-only \
  "Review this PR diff. Focus on: bugs, edge cases, missing error handling, 
   race conditions, security issues. Be specific — cite file and line numbers.
   Output format: list of issues with severity (critical/warning/info)."

# Option B: Claude reviews with security focus
echo "$DIFF" | claude -p \
  --append-system-prompt "You are a security-focused code reviewer. Flag only critical issues." \
  "Review this diff for security vulnerabilities, injection risks, and logic errors."

Post review results to PR:

gh pr comment "$PR_NUM" --body "## AI Code Review

$REVIEW_OUTPUT"

Update task registry:

(
  flock -x 200
  jq --arg id "$TASK_ID" \
    '(.tasks[] | select(.id == $id)).checks.codeReviewPassed = true' \
    "$TASKS_FILE" > /tmp/tasks.$$.json && mv /tmp/tasks.$$.json "$TASKS_FILE"
) 200>"$TASKS_FILE.lock"

Step 7: Notify

If you included the notify command in the agent prompt (Step 2), the agent self-notifies on completion.

Otherwise, notify after DoD passes:

openclaw system event --text "✅ PR #$PR_NUM ready for review: $TASK_ID — all checks passed" --mode now

Update task status:

(
  flock -x 200
  jq --arg id "$TASK_ID" --argjson pr "$PR_NUM" \
    '(.tasks[] | select(.id == $id)) |= (.status = "done" | .pr = $pr | .completedAt = (now|floor))' \
    "$TASKS_FILE" > /tmp/tasks.$$.json && mv /tmp/tasks.$$.json "$TASKS_FILE"
) 200>"$TASKS_FILE.lock"

Step 8: Cleanup

After PR is merged:

git worktree remove "$WORKTREE" 2>/dev/null
git branch -d "$BRANCH" 2>/dev/null

# Remove from registry
(
  flock -x 200
  jq --arg id "$TASK_ID" '.tasks = [.tasks[] | select(.id != $id)]' \
    "$TASKS_FILE" > /tmp/tasks.$$.json && mv /tmp/tasks.$$.json "$TASKS_FILE"
) 200>"$TASKS_FILE.lock"

Smart Retry Strategy

When an agent fails, analyze the failure and adapt the prompt — don't just re-run blindly.

Max 3 retries. After that, escalate to human.

RETRY=$((RETRY + 1))
if [ "$RETRY" -gt 3 ]; then
  openclaw system event --text "BLOCKED: $TASK_ID failed after 3 retries — needs human help" --mode now
  (
    flock -x 200
    jq --arg id "$TASK_ID" '(.tasks[] | select(.id == $id)).status = "blocked"' \
      "$TASKS_FILE" > /tmp/tasks.$$.json && mv /tmp/tasks.$$.json "$TASKS_FILE"
  ) 200>"$TASKS_FILE.lock"
  exit 1
fi

# Capture what went wrong — prefer log file over tmux scrollback
LOG_FILE="/tmp/worktrees/$TASK_ID/claude-output.log"
if [ -f "$LOG_FILE" ]; then
  # Log file is ANSI-stripped (clean text) — extract error-relevant lines
  # Take last 500 lines, but also grep for error context
  FAILURE_LOG=$(tail -500 "$LOG_FILE")
  ERROR_LINES=$(grep -n -i "error\|fail\|panic\|exception\|traceback" "$LOG_FILE" | tail -50)
  if [ -n "$ERROR_LINES" ]; then
    FAILURE_LOG="=== Error lines ===
$ERROR_LINES

=== Last 500 lines ===
$FAILURE_LOG"
  fi
else
  FAILURE_LOG=$(tmux capture-pane -t "$TASK_ID" -p -S -200)
fi
CI_LOG=$(gh pr checks "$PR_NUM" 2>/dev/null || echo "no PR yet")
tmux kill-session -t "$TASK_ID" 2>/dev/null

# Archive old log, start fresh for retry
[ -f "$LOG_FILE" ] && mv "$LOG_FILE" "${LOG_FILE}.retry$((RETRY - 1))"

# Respawn — set up pipe-pane BEFORE sending command (captures all output)
tmux new-session -d -s "$TASK_ID" -c "$WORKTREE"
tmux pipe-pane -t "$TASK_ID" -o "sed 's/\x1b\[[0-9;]*[a-zA-Z]//g' >> $LOG_FILE"
tmux send-keys -t "$TASK_ID" "claude --dangerously-skip-permissions \
  'Previous attempt failed. Error output:
$FAILURE_LOG

CI status: $CI_LOG

Fix the issues above and complete the original task.
[...your enriched instructions here...]

When done: commit, push, gh pr create --fill, then run:
openclaw system event --text \"Done: $TASK_ID (retry $RETRY)\" --mode now'" Enter

# Update registry with flock
PANE_PID=$(tmux display-message -t "$TASK_ID" -p '#{pane_pid}')
(
  flock -x 200
  jq --arg id "$TASK_ID" --argjson r "$RETRY" --arg pane_pid "$PANE_PID" \
    '(.tasks[] | select(.id == $id)) |= (.retries = $r | .status = "running" | .panePid = ($pane_pid | tonumber) | .silentRounds = 0 | .repeatingRounds = 0 | .lastOutputHash = "")' \
    "$TASKS_FILE" > /tmp/tasks.$$.json && mv /tmp/tasks.$$.json "$TASKS_FILE"
) 200>"$TASKS_FILE.lock"

Parallel Execution

Run multiple agents simultaneously on different tasks. Important: Always set up pipe-pane before sending the command to avoid losing early output. Use ANSI stripping for clean logs.

# Helper: launch an agent in tmux with proper output capture
launch_agent() {
  local TASK_ID="$1" WORKTREE="$2" PROMPT="$3"
  local LOG_FILE="$WORKTREE/claude-output.log"

  # 1. Create session with idle shell
  tmux new-session -d -s "$TASK_ID" -c "$WORKTREE"
  # 2. Start pipe-pane BEFORE command (captures everything, ANSI-stripped)
  tmux pipe-pane -t "$TASK_ID" -o "sed 's/\x1b\[[0-9;]*[a-zA-Z]//g' >> $LOG_FILE"
  # 3. Send command
  tmux send-keys -t "$TASK_ID" "$PROMPT" Enter
}

# Task 1: Feature
git worktree add -b feat/auth /tmp/worktrees/feat-auth origin/main
launch_agent "feat-auth" "/tmp/worktrees/feat-auth" \
  "cd /tmp/worktrees/feat-auth && pnpm install && claude --dangerously-skip-permissions 'Implement JWT auth...'"

# Task 2: Bugfix
git worktree add -b fix/payments /tmp/worktrees/fix-payments origin/main
launch_agent "fix-payments" "/tmp/worktrees/fix-payments" \
  "cd /tmp/worktrees/fix-payments && pnpm install && claude --dangerously-skip-permissions 'Fix payment webhook...'"

# Dashboard: check all agents (uses process check, not just has-session)
echo "=== Agent Status ==="
for s in $(tmux ls -F '#{session_name}' 2>/dev/null); do
  PANE_PID=$(tmux display-message -t "$s" -p '#{pane_pid}' 2>/dev/null)
  if [ -z "$PANE_PID" ]; then
    STATUS="(session gone)"
  elif pgrep -P "$PANE_PID" > /dev/null 2>&1; then
    STATUS="running"
  else
    STATUS="process exited"
  fi
  LOG="/tmp/worktrees/$s/claude-output.log"
  LAST_LINE=$(tail -1 "$LOG" 2>/dev/null || echo "(no log)")
  echo "  $s: $STATUS | last: $LAST_LINE"
done

Multi-Turn Conversations

For complex tasks that need iterative refinement:

# Start session, capture ID
session_id=$(claude -p "analyze the codebase architecture" \
  --output-format json | jq -r '.session_id')

# Continue with context from previous turn
claude -r "$session_id" -p "now implement the changes we discussed" \
  --dangerously-skip-permissions

# Resume in non-interactive mode
claude -r "$session_id" -p "fix the remaining test failures" \
  --dangerously-skip-permissions

Best Practices

Prompt Quality

• Include file paths: "The entry point is src/index.ts, config in src/config/"
• Include schemas/types: Paste relevant type definitions into the prompt
• Include test commands: "Verify with: npm test -- --grep auth"
• Include commit convention: "Use conventional commits: feat:, fix:, chore:"
• Include error logs: When retrying, always attach the failure output

Scope Management

• One task per agent — don't ask for "refactor everything"
• Pre-install dependencies before launching the agent
• Be specific — "Add rate limiting to POST /api/users" not "improve the API"

When to Interrupt (Ask Human)

• Destructive operations (drop tables, force push main)
• Security decisions (expose credentials, change auth)
• Ambiguous requirements with significant trade-offs
• All other decisions: proceed autonomously

Examples

See references/examples.md for additional usage scenarios.