subagent

Execute one or more issues as subagents within the current thread. Each issue goes through the full Phase A/B/C orchestration protocol (fix, review, merge prep) while this skill monitors progress and manages transitions. Inline execution is the default for issues of any tier. Only two of Step 4's three too-big criteria route an issue out to a separate thread; the third — "should be split into multiple PRs" — is decomposed into an inline increment chain instead (Step 5.1).

Parse $ARGUMENTS as space-separated issue references. Strip # prefixes to get bare issue numbers. If no arguments provided, ask the user which issue(s) to execute.

---

Step 0: Resolve shared tooling

/subagent is symlinked into every repo, but its helper scripts and reference docs are not — most repos carry no .claude/ directory. Resolve them; never invoke a bare .claude/scripts/… path. Full contract and the classified dependency inventory: .claude/reference/portable-skill-resolution.md (issue #1189).

resolve_script() {
  local name="$1" candidate
  for candidate in \
    "$HOME/.claude/skills-worktree/.claude/scripts/$name" \
    "$HOME/.claude/scripts/$name" \
    ".claude/scripts/$name"; do
    if [[ -x "$candidate" ]]; then echo "$candidate"; return 0; fi
  done
  return 1
}
ISSUE_CLAIM=$(resolve_script issue-claim.sh || true)
CR_PLAN=$(resolve_script cr-plan.sh || true)
SESSION_STATE_SH=$(resolve_script session-state.sh || true)
ISSUE_DEDUP=$(resolve_script issue-dedup.sh || true)
ESTIMATE_RESOLVE_SH=$(resolve_script estimate-resolve.sh || true)
OVERRUN_CHECK_SH=$(resolve_script overrun-check.sh || true)

handoff-state.sh (Step 8), ac-checkboxes.sh, escalate-review.sh, and local-review.sh are resolved by the phase agents themselves, inside the spawn prompts — the RESOLVE block inserted with SAFETY/MINDSET/SKILLS carries the same candidate order to them. Read reference docs (chip-launching.md, subagent-phase-guardrails.md, issue-claim.md, merge-sequencing.md) through the matching .claude/reference/ order.

When something does not resolve, say so in one line; never skip the contract silently.

• subagent-phase-guardrails.md unreadable → required. Print ERROR: subagent-phase-guardrails.md not found (checked all three paths) — SAFETY/MINDSET/SKILLS/RESOLVE blocks unavailable and spawn nothing. Those blocks are the safety restatement every spawn carries; a phase agent launched without them is not a degraded spawn, it is an unsafe one.
• chip-launching.md unreadable → required for Step 5's route-to-thread branch only. Print ERROR: chip-launching.md not found (checked all three paths) — too-big routing gate unavailable, and queue the issue inline rather than routing it out: inline is the default the gate exists to protect (#1189), so failing toward it is correct.
• SESSION_STATE_SH empty → required. Print ERROR: session-state.sh not found (checked all three paths) — agent tracking and refill state unavailable and do not spawn; an untracked agent is one nothing will ever reap.
• ISSUE_CLAIM empty → optional. Print DEGRADED: issue-claim.sh not found (checked all three paths) — Step 6 claim gate skipped and continue.
• CR_PLAN empty → optional. Print DEGRADED: cr-plan.sh not found (checked all three paths) — CR plan detection skipped and continue with Claude's own plan.
• ESTIMATE_RESOLVE_SH empty → optional. Print DEGRADED: estimate-resolve.sh not found (checked all three paths) — planning-bound lookup unavailable; overrun check skipped and skip the overrun check in Step 8 (BOUND_MIN cannot be derived without it).
• OVERRUN_CHECK_SH empty → optional. Print DEGRADED: overrun-check.sh not found (checked all three paths) — in-flight overrun alerts unavailable and skip the overrun check in Step 8.

The Step 4 too-big criteria need no fallback — they are written inline in this file, so that contract already travels. Only their per-criterion rationale doc (too-big-recalibration-2026-07.md) is a fallback read.