claude-code-config: Skill for Claude Code

You say the time once. From then on the thread plans around it: nothing starts that cannot finish before it, a check-in arrives at a lead time ahead of it, and the wind-down runs itself so that by the declared time everything is merged or cleanly resumable.

This skill is a thin layer over three mechanisms that already exist. It arms /pm Step 0b's planning window (issue #1325) at the declared time, it schedules /pause (issue #1482) to run at deadline − lead, and it renders /subagent's "Running now" table (issue #1512) with one added verdict column. It is not a second pause implementation, a second deadline field, or a second scheduler — every one of those would be a place for the two copies to disagree.

Arming a leave time does not turn this thread into a /pm thread. It writes repo-scoped state that any execution-capable thread reads; it imports no ranking, no backlog scan, and no day loop.

Step 0: Resolve helpers, repo key, and mode

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
}
SESSION_STATE_SH=$(resolve_script session-state.sh)     || SESSION_STATE_SH=""
WINDOW_PLAN_SH=$(resolve_script window-plan.sh)         || WINDOW_PLAN_SH=""
PM_CONFIG_GET=$(resolve_script pm-config-get.sh)        || PM_CONFIG_GET=""
ESTIMATE_RESOLVE_SH=$(resolve_script estimate-resolve.sh) || ESTIMATE_RESOLVE_SH=""
OVERRUN_CHECK_SH=$(resolve_script overrun-check.sh)     || OVERRUN_CHECK_SH=""
TABLE_FRESHNESS_SH=$(resolve_script table-freshness.sh) || TABLE_FRESHNESS_SH=""

REPO_KEY=""
[[ -n "$SESSION_STATE_SH" ]] && { REPO_KEY=$("$SESSION_STATE_SH" --repo-key 2>/dev/null) || REPO_KEY=""; }
# `--repo-key` NEVER returns empty: it prints `_unknown` and exits 0 when no
# repo resolves, so the empty-REPO_KEY branch below cannot fire without this.
# Left unnormalised, the check-in's `--note-rendered` writes under `_unknown`
# while the armed watch polls the real repo — the render is recorded where
# nothing looks, and the floor fires forever against a board being re-rendered.
[[ "$REPO_KEY" == "_unknown" ]] && REPO_KEY=""

• session-state.sh or window-plan.sh unresolved, or an empty REPO_KEY → required. Print ERROR: <name> not found (checked all three paths) — leave-time arming unavailable (or ERROR: repo key unresolved — leave-time arming unavailable) and stop. A leave time that is not persisted is a promise nothing will keep; refusing is the honest failure.
• pm-config-get.sh unresolved → degraded: DEGRADED: pm-config-get.sh not found (checked all three paths) — lead time falls back to 30 min, then continue.
• table-freshness.sh unresolved → degraded: DEGRADED: table-freshness.sh not found (checked all three paths) — hourly table-freshness floor unavailable; re-render the "Running now" table on every heartbeat instead, then continue. Same wording and same direction as /subagent Step 0: the check-in still prints (Step 8.3), it just goes unrecorded, and failing toward more table renders is correct.
• estimate-resolve.sh / overrun-check.sh unresolved → degraded: the check-in table loses its Est and clock columns (unestimated / —), and every started row's verdict falls to parks (Step 8). Say so in one line; never skip the check-in.

Modes, decided before anything else. Parse the internal fields first so a Monitor-emitted event can never be reinterpreted as a fresh declaration: