agent-control-plane 0.1.14 → 0.2.0

package/tools/bin/agent-project-run-claude-session

@@ -774,12 +774,12 @@ fi
 
 ${collect_copy_snippet}
 if [[ "\${status}" -eq 0 ]]; then
-  write_state completed "\${status}" '' "\${attempt}" "\$((attempt - 1))"
+  write_state succeeded "\${status}" '' "\${attempt}" "\$((attempt - 1))"
 else
   write_state failed "\${status}" "\${failure_reason}" "\${attempt}" "\$((attempt - 1))"
 fi
 ${reconcile_snippet}
-printf '\n__CLAUDE_EXIT__:%s\n' "\${status}" | tee -a "\${output_file}"
+printf '\n__CODEX_EXIT__:%s\n' "\${status}" | tee -a "\${output_file}"
 exit "\${status}"
 EOF
 

package/tools/bin/agent-project-run-codex-resilient

@@ -618,6 +618,22 @@ classify_failure_reason() {
   fi
 }
 
+failure_chunk_indicates_startup_stall() {
+  local chunk="${1:-}"
+  local recent_chunk
+
+  recent_chunk="$(tail -n 120 <<<"$chunk")"
+  grep -q '"type":"thread.started"' <<<"$recent_chunk" || return 1
+  grep -q '"type":"turn.started"' <<<"$recent_chunk" || return 1
+  if grep -Eq '"type":"item\.(started|completed)"' <<<"$recent_chunk"; then
+    return 1
+  fi
+  if grep -q '"type":"turn.completed"' <<<"$recent_chunk"; then
+    return 1
+  fi
+  return 0
+}
+
 resume_prompt() {
   cat <<EOF
 The previous Codex exec turn in this same thread was interrupted because the host refreshed Codex authentication after a quota or auth failure.
@@ -729,7 +745,7 @@ run_resume_exec() {
 }
 
 attempt_run() {
-  local reason auth_before_switch quota_label_before_switch quota_switch_signature_before_switch quota_switch_result shell_flags_before_quota_switch
+  local reason auth_before_switch quota_label_before_switch quota_switch_signature_before_switch quota_switch_result shell_flags_before_quota_switch failure_chunk startup_stall
 
   attempt=$((attempt + 1))
   last_quota_switch_status=""
@@ -750,8 +766,15 @@ attempt_run() {
     return 0
   fi
 
-  reason="$(classify_failure_reason "$(new_output_since "$last_attempt_start_size")")"
+  failure_chunk="$(new_output_since "$last_attempt_start_size")"
+  reason="$(classify_failure_reason "$failure_chunk")"
   last_failure_reason="${reason:-worker-exit-failed}"
+  startup_stall="no"
+  if [[ "$last_failure_reason" == "no-codex-output-before-stall-threshold" || "$last_failure_reason" == "no-codex-progress-before-stall-threshold" ]]; then
+    if failure_chunk_indicates_startup_stall "$failure_chunk"; then
+      startup_stall="yes"
+    fi
+  fi
 
   case "$last_failure_reason" in
     usage-limit|auth-failure|auth-401|account-banned)
@@ -796,6 +819,38 @@ attempt_run() {
       resume_count=$((resume_count + 1))
       return 2
       ;;
+    no-codex-output-before-stall-threshold|no-codex-progress-before-stall-threshold)
+      if [[ "$startup_stall" == "yes" && $quota_autoswitch_attempt_count -lt $max_quota_autoswitch_attempts ]]; then
+        auth_before_switch="$(auth_fingerprint)"
+        quota_label_before_switch="$last_attempt_start_quota_label"
+        quota_switch_signature_before_switch="$(quota_switch_signature)"
+        last_auth_fingerprint="$auth_before_switch"
+        write_state "switching-account" "$last_failure_reason"
+        log_runner "startup-stall detected before first Codex tool activity; attempting Codex account rotation"
+        shell_flags_before_quota_switch="$-"
+        set +e
+        run_quota_autoswitch
+        quota_switch_result=$?
+        case "$shell_flags_before_quota_switch" in
+          *e*) set -e ;;
+          *) set +e ;;
+        esac
+        if [[ "$quota_switch_result" == "0" ]]; then
+          thread_id=""
+          auth_wait_started_at=""
+          write_state "running" ""
+          return 2
+        fi
+        if [[ "$quota_switch_result" == "10" ]]; then
+          log_runner "startup-stall rotation deferred until ${last_quota_next_retry_at:-unknown}"
+          last_failure_reason="quota-switch-deferred"
+          write_state "failed" "$last_failure_reason"
+          return 1
+        fi
+      fi
+      write_state "failed" "$last_failure_reason"
+      return 1
+      ;;
     *)
       write_state "failed" "$last_failure_reason"
       return 1

package/tools/bin/agent-project-run-kilo-session

@@ -4,24 +4,356 @@ set -euo pipefail
 usage() {
   cat <<'EOF'
 Usage:
-  agent-project-run-kilo-session [shared session-wrapper options]
+  agent-project-run-kilo-session --mode safe|bypass --session <id> --worktree <path> --prompt-file <path> --runs-root <path> --adapter-id <id> --task-kind <kind> --task-id <id> [options]
 
-Placeholder adapter for the roadmap. The public package ships this command so
-ACP can expose and test worker routing for `kilo`, but live session execution
-is not implemented yet.
+Launch a Kilo Code worker session inside tmux for a project adapter and persist
+the standard run artifacts.
 
-Use `codex`, `claude`, or `openclaw` for real runs today. See ROADMAP.md for
-current backend maturity.
+Kilo is a TypeScript/Bun coding agent (kilocode/cli). It executes via
+`kilo run --auto --format json` in non-interactive mode.
+
+Options:
+  --env-prefix <prefix>             Export prefixed runtime/context env vars inside the worker
+  --context <KEY=VALUE>             Extra metadata written to run.env and exported to the worker
+  --collect-file <name>             Copy sandbox artifact file into the host run dir after execution
+  --reconcile-command <cmd>         Host-side command queued after the worker exits
+  --sandbox-subdir <name>           Subdir under the worktree for worker artifacts (default: .kilo-artifacts)
+  --kilo-model <id>                 Model in provider/name format (default: anthropic/claude-sonnet-4-20250514)
+  --kilo-timeout-seconds <secs>     Hard timeout in seconds (default: 900)
+  --help                            Show this help
 EOF
 }
 
-case "${1:-}" in
-  --help|-h)
-    usage
-    exit 0
-    ;;
+mode=""
+session=""
+worktree=""
+prompt_file=""
+runs_root=""
+adapter_id=""
+task_kind=""
+task_id=""
+env_prefix=""
+sandbox_subdir=".kilo-artifacts"
+reconcile_command=""
+kilo_model="${ACP_KILO_MODEL:-${F_LOSNING_KILO_MODEL:-anthropic/claude-sonnet-4-20250514}}"
+kilo_timeout_seconds="${ACP_KILO_TIMEOUT_SECONDS:-${F_LOSNING_KILO_TIMEOUT_SECONDS:-900}}"
+declare -a context_items=()
+declare -a collect_files=()
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --mode) mode="${2:-}"; shift 2 ;;
+    --session) session="${2:-}"; shift 2 ;;
+    --worktree) worktree="${2:-}"; shift 2 ;;
+    --prompt-file) prompt_file="${2:-}"; shift 2 ;;
+    --runs-root) runs_root="${2:-}"; shift 2 ;;
+    --adapter-id) adapter_id="${2:-}"; shift 2 ;;
+    --task-kind) task_kind="${2:-}"; shift 2 ;;
+    --task-id) task_id="${2:-}"; shift 2 ;;
+    --env-prefix) env_prefix="${2:-}"; shift 2 ;;
+    --context) context_items+=("${2:-}"); shift 2 ;;
+    --collect-file) collect_files+=("${2:-}"); shift 2 ;;
+    --reconcile-command) reconcile_command="${2:-}"; shift 2 ;;
+    --sandbox-subdir) sandbox_subdir="${2:-}"; shift 2 ;;
+    --kilo-model) kilo_model="${2:-}"; shift 2 ;;
+    --kilo-timeout-seconds) kilo_timeout_seconds="${2:-}"; shift 2 ;;
+    --help|-h) usage; exit 0 ;;
+    *) echo "Unknown argument: $1" >&2; usage >&2; exit 1 ;;
+  esac
+done
+
+if [[ -z "$mode" || -z "$session" || -z "$worktree" || -z "$prompt_file" || -z "$runs_root" || -z "$adapter_id" || -z "$task_kind" || -z "$task_id" ]]; then
+  usage >&2
+  exit 1
+fi
+
+case "$mode" in
+  safe|bypass) ;;
+  *) echo "--mode must be safe or bypass" >&2; exit 1 ;;
+esac
+
+case "$kilo_timeout_seconds" in
+  ''|*[!0-9]*|0) echo "--kilo-timeout-seconds must be a positive integer" >&2; exit 1 ;;
 esac
 
-echo "agent-project-run-kilo-session: kilo support is scaffolded, but execution is not implemented yet" >&2
-echo "Choose codex, claude, or openclaw for live runs today." >&2
-exit 1
+resolve_kilo_bin() {
+  local configured_bin="${KILO_BIN:-${ACP_KILO_BIN:-}}"
+  if [[ -n "${configured_bin}" && -x "${configured_bin}" ]]; then
+    printf '%s\n' "${configured_bin}"
+    return 0
+  fi
+  if command -v kilo >/dev/null 2>&1; then
+    command -v kilo
+    return 0
+  fi
+  local -a fallback_paths=(
+    "/opt/homebrew/bin/kilo"
+    "/usr/local/bin/kilo"
+    "${HOME}/.local/bin/kilo"
+    "${HOME}/.bun/bin/kilo"
+  )
+  local p
+  for p in "${fallback_paths[@]}"; do
+    if [[ -x "${p}" ]]; then
+      printf '%s\n' "${p}"
+      return 0
+    fi
+  done
+  return 1
+}
+
+kilo_bin="$(resolve_kilo_bin || true)"
+if [[ -z "${kilo_bin}" || ! -x "${kilo_bin}" ]]; then
+  echo "unable to resolve a runnable kilo binary — install with: npm install -g @kilocode/cli" >&2
+  exit 1
+fi
+
+artifact_dir="${runs_root}/${session}"
+output_file="${artifact_dir}/${session}.log"
+inner_script="${artifact_dir}/${session}.sh"
+meta_file="${artifact_dir}/run.env"
+result_file="${artifact_dir}/result.env"
+runner_state_file="${artifact_dir}/runner.env"
+sandbox_run_dir="${worktree%/}/${sandbox_subdir}/${session}"
+started_at="$(date -u +"%Y-%m-%dT%H:%M:%SZ")"
+
+mkdir -p "$artifact_dir" "$sandbox_run_dir"
+
+if tmux has-session -t "$session" 2>/dev/null; then
+  echo "tmux session already exists: $session" >&2
+  exit 1
+fi
+
+branch_name="$(git -C "$worktree" branch --show-current 2>/dev/null || true)"
+
+printf -v session_q '%q' "$session"
+printf -v task_kind_q '%q' "$task_kind"
+printf -v task_id_q '%q' "$task_id"
+printf -v mode_q '%q' "$mode"
+printf -v worktree_q '%q' "$worktree"
+printf -v prompt_q '%q' "$prompt_file"
+printf -v output_q '%q' "$output_file"
+printf -v artifact_dir_q '%q' "$artifact_dir"
+printf -v script_q '%q' "$inner_script"
+printf -v result_q '%q' "$result_file"
+printf -v meta_file_q '%q' "$meta_file"
+printf -v runner_state_q '%q' "$runner_state_file"
+printf -v branch_q '%q' "$branch_name"
+printf -v sandbox_run_dir_q '%q' "$sandbox_run_dir"
+printf -v adapter_id_q '%q' "$adapter_id"
+printf -v started_at_q '%q' "$started_at"
+printf -v kilo_bin_q '%q' "$kilo_bin"
+printf -v kilo_model_q '%q' "$kilo_model"
+printf -v kilo_timeout_q '%q' "$kilo_timeout_seconds"
+
+{
+  printf 'TASK_KIND=%s\n' "$task_kind_q"
+  printf 'TASK_ID=%s\n' "$task_id_q"
+  printf 'SESSION=%s\n' "$session_q"
+  printf 'MODE=%s\n' "$mode_q"
+  printf 'WORKTREE=%s\n' "$worktree_q"
+  printf 'PROMPT_FILE=%s\n' "$prompt_q"
+  printf 'OUTPUT_FILE=%s\n' "$output_q"
+  printf 'BRANCH=%s\n' "$branch_q"
+  printf 'RESULT_FILE=%s\n' "$result_q"
+  printf 'RUNNER_STATE_FILE=%s\n' "$runner_state_q"
+  printf 'SANDBOX_RUN_DIR=%s\n' "$sandbox_run_dir_q"
+  printf 'ADAPTER_ID=%s\n' "$adapter_id_q"
+  printf 'STARTED_AT=%s\n' "$started_at_q"
+  printf 'KILO_BIN=%s\n' "$kilo_bin_q"
+  printf 'KILO_MODEL=%s\n' "$kilo_model_q"
+  printf 'KILO_TIMEOUT_SECONDS=%s\n' "$kilo_timeout_q"
+} >"$meta_file"
+
+context_exports=""
+if ((${#context_items[@]} > 0)); then
+  for item in "${context_items[@]}"; do
+    if [[ "$item" != *=* ]]; then
+      echo "--context must use KEY=VALUE syntax: $item" >&2
+      exit 1
+    fi
+    key="${item%%=*}"
+    value="${item#*=}"
+    if [[ ! "$key" =~ ^[A-Z0-9_]+$ ]]; then
+      echo "Invalid context key: $key" >&2
+      exit 1
+    fi
+    printf -v value_q '%q' "$value"
+    printf '%s=%s\n' "$key" "$value_q" >>"$meta_file"
+    if [[ -n "$env_prefix" ]]; then
+      context_exports+="export ${env_prefix}${key}=${value_q}"$'\n'
+    fi
+    context_exports+="export ACP_${key}=${value_q}"$'\n'
+    if [[ "$env_prefix" != "F_LOSNING_" ]]; then
+      context_exports+="export F_LOSNING_${key}=${value_q}"$'\n'
+    fi
+  done
+fi
+
+runtime_exports=$(
+  cat <<EOF
+export AGENT_PROJECT_SESSION=${session_q}
+export AGENT_PROJECT_RUN_DIR=${sandbox_run_dir_q}
+export AGENT_PROJECT_HOST_RUN_DIR=${artifact_dir_q}
+export AGENT_PROJECT_RESULT_FILE=${sandbox_run_dir_q}/result.env
+export ACP_SESSION=${session_q}
+export ACP_RUN_DIR=${sandbox_run_dir_q}
+export ACP_HOST_RUN_DIR=${artifact_dir_q}
+export ACP_RESULT_FILE=${sandbox_run_dir_q}/result.env
+export F_LOSNING_SESSION=${session_q}
+export F_LOSNING_RUN_DIR=${sandbox_run_dir_q}
+export F_LOSNING_HOST_RUN_DIR=${artifact_dir_q}
+export F_LOSNING_RESULT_FILE=${sandbox_run_dir_q}/result.env
+EOF
+)
+
+if [[ -n "$env_prefix" ]]; then
+  runtime_exports+=$'\n'
+  runtime_exports+=$(cat <<EOF
+export ${env_prefix}SESSION=${session_q}
+export ${env_prefix}RUN_DIR=${sandbox_run_dir_q}
+export ${env_prefix}HOST_RUN_DIR=${artifact_dir_q}
+export ${env_prefix}RESULT_FILE=${sandbox_run_dir_q}/result.env
+EOF
+)
+fi
+
+collect_copy_snippet=""
+if ((${#collect_files[@]} > 0)); then
+  for artifact_name in "${collect_files[@]}"; do
+    [[ -z "$artifact_name" ]] && continue
+    printf -v artifact_q '%q' "$artifact_name"
+    collect_copy_snippet+=$(cat <<EOF
+if [[ -f ${sandbox_run_dir_q}/${artifact_q} ]]; then
+  cp ${sandbox_run_dir_q}/${artifact_q} ${artifact_dir_q}/${artifact_q}
+fi
+EOF
+    )
+    collect_copy_snippet+=$'\n'
+  done
+fi
+
+reconcile_snippet=""
+if [[ -n "$reconcile_command" ]]; then
+  printf -v delayed_reconcile_q '%q' "export ACP_EXPECTED_RUN_STARTED_AT=${started_at_q}; export F_LOSNING_EXPECTED_RUN_STARTED_AT=${started_at_q}; while tmux has-session -t ${session_q} 2>/dev/null; do sleep 1; done; sleep 2; $reconcile_command"
+  reconcile_snippet="nohup bash -lc ${delayed_reconcile_q} >> ${output_q} 2>&1 </dev/null &"
+fi
+
+cat >"$inner_script" <<EOF
+#!/usr/bin/env bash
+set -euo pipefail
+${runtime_exports}
+${context_exports}cd ${worktree_q}
+
+runner_state_file=${runner_state_q}
+output_file=${output_q}
+sandbox_run_dir=${sandbox_run_dir_q}
+artifact_dir=${artifact_dir_q}
+result_file_path=${sandbox_run_dir_q}/result.env
+host_result_file=${result_q}
+kilo_bin=${kilo_bin_q}
+kilo_model=${kilo_model_q}
+kilo_timeout=${kilo_timeout_q}
+prompt_file=${prompt_q}
+worktree=${worktree_q}
+
+write_state() {
+  local runner_state="\${1:?runner state required}"
+  local last_exit_code="\${2:-}"
+  local failure_reason="\${3:-}"
+  local updated_at tmp_file
+
+  updated_at="\$(date -u +"%Y-%m-%dT%H:%M:%SZ")"
+  tmp_file="\${runner_state_file}.tmp.\$\$"
+  {
+    printf 'RUNNER_STATE=%q\n' "\${runner_state}"
+    printf 'ATTEMPT=1\n'
+    printf 'RESUME_COUNT=0\n'
+    printf 'LAST_EXIT_CODE=%q\n' "\${last_exit_code}"
+    printf 'LAST_FAILURE_REASON=%q\n' "\${failure_reason}"
+    printf 'UPDATED_AT=%q\n' "\${updated_at}"
+  } >"\${tmp_file}"
+  mv "\${tmp_file}" "\${runner_state_file}"
+}
+
+record_final_git_state() {
+  local final_head final_branch tmp_file
+  final_head="\$(git -C ${worktree_q} rev-parse HEAD 2>/dev/null || true)"
+  final_branch="\$(git -C ${worktree_q} branch --show-current 2>/dev/null || true)"
+  tmp_file=${meta_file_q}.tmp.final.\$\$
+  grep -vE '^(FINAL_HEAD|FINAL_BRANCH)=' ${meta_file_q} >"\${tmp_file}" 2>/dev/null || true
+  {
+    printf 'FINAL_HEAD=%q\n' "\${final_head}"
+    printf 'FINAL_BRANCH=%q\n' "\${final_branch}"
+  } >>"\${tmp_file}"
+  mv "\${tmp_file}" ${meta_file_q}
+}
+
+write_state running
+
+mkdir -p "\${sandbox_run_dir}"
+
+# Kilo runs via 'kilo run' in non-interactive mode.
+# --auto auto-approves all tool permissions (CI mode).
+# --format json emits structured JSON events (parseable for result inference).
+# --model selects the provider/model, --dir sets the working directory.
+# Prompt is passed as the positional argument.
+prompt_content="\$(cat "\${prompt_file}")"
+kilo_exit_code=0
+kilo_args=(run --auto --format json --model "\${kilo_model}" --dir ${worktree_q})
+
+if command -v timeout >/dev/null 2>&1; then
+  timeout "\${kilo_timeout}" "\${kilo_bin}" "\${kilo_args[@]}" "\${prompt_content}" 2>&1 | tee -a "\${output_file}" || kilo_exit_code=\$?
+elif command -v gtimeout >/dev/null 2>&1; then
+  gtimeout "\${kilo_timeout}" "\${kilo_bin}" "\${kilo_args[@]}" "\${prompt_content}" 2>&1 | tee -a "\${output_file}" || kilo_exit_code=\$?
+else
+  "\${kilo_bin}" "\${kilo_args[@]}" "\${prompt_content}" 2>&1 | tee -a "\${output_file}" &
+  _kilo_pid=\$!
+  ( sleep "\${kilo_timeout}" && kill "\${_kilo_pid}" 2>/dev/null ) &
+  _wd=\$!
+  wait "\${_kilo_pid}" || kilo_exit_code=\$?
+  kill "\${_wd}" 2>/dev/null || true; wait "\${_wd}" 2>/dev/null || true
+  [[ "\${kilo_exit_code}" -eq 143 ]] && kilo_exit_code=124
+fi
+
+if [[ "\${kilo_exit_code}" -eq 0 ]]; then
+  write_state succeeded 0
+  # Kilo has full tool access (bash, write, edit) — it can modify the worktree.
+  # Infer result from git state if result.env was not written by the agent.
+  if [[ ! -f "\${result_file_path}" ]]; then
+    if git -C ${worktree_q} diff --name-only HEAD 2>/dev/null | grep -qvE '\.md$' 2>/dev/null \\
+      || git -C ${worktree_q} diff --cached --name-only 2>/dev/null | grep -qvE '\.md$' 2>/dev/null \\
+      || git -C ${worktree_q} diff --name-only origin/main..HEAD 2>/dev/null | grep -qvE '\.md$' 2>/dev/null; then
+      printf 'OUTCOME=implemented\nACTION=host-publish-issue-pr\n' >"\${result_file_path}"
+    else
+      printf 'OUTCOME=blocked\nACTION=host-comment-blocker\nDETAIL=missing-result-contract\n' >"\${result_file_path}"
+    fi
+  fi
+else
+  failure_reason="kilo-exit-\${kilo_exit_code}"
+  [[ "\${kilo_exit_code}" -eq 124 ]] && failure_reason="timeout"
+  write_state failed "\${kilo_exit_code}" "\${failure_reason}"
+  if [[ ! -f "\${result_file_path}" ]]; then
+    printf 'OUTCOME=blocked\nACTION=host-comment-blocker\nDETAIL=%s\n' "\${failure_reason}" >"\${result_file_path}"
+  fi
+fi
+
+record_final_git_state
+
+if [[ -f "\${result_file_path}" ]]; then
+  cp "\${result_file_path}" "\${host_result_file}"
+fi
+${collect_copy_snippet}
+${reconcile_snippet}
+printf '\n__CODEX_EXIT__:%s\n' "\${kilo_exit_code}" | tee -a "\${output_file}"
+exit "\${kilo_exit_code}"
+EOF
+
+chmod +x "$inner_script"
+tmux new-session -d -s "$session" "$inner_script"
+
+printf 'SESSION=%s\n' "$session"
+printf 'TASK_KIND=%s\n' "$task_kind"
+printf 'TASK_ID=%s\n' "$task_id"
+printf 'WORKTREE=%s\n' "$worktree"
+printf 'OUTPUT=%s\n' "$output_file"