@delorenj/pjangler 1.2.25 → 1.3.0

package/templates/hermes-agent/scripts/migrate-unify.sh

@@ -1,252 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-# migrate-unify — one-time fleet migration to the unified single-PM model.
-#
-# Decommissions the legacy `scrum-master` agents, collapsing each repo from a
-# PM + scrum-master pair down to a single PM. The PM keeps running untouched;
-# its checkpoint→heartbeat refresh is a separate uniform pass (and with
-# reconcile disabled by default the PM heartbeat is checkpoint-only, so removing
-# the scrum-master is behavior-complete for the collapse).
-#
-# Drives ENTIRELY off the registry's role=scrum-master entries
-# (role_dir / project_path / runtime_repo) — NEVER name-derived paths, because
-# keepy-money lives at ~/code/tiller, zshyzsh at ~/.config/zshyzsh, and
-# delocontainers at ~/docker.
-#
-# Safety model:
-#   * DRY-RUN by default — prints exactly what it WOULD do, touches nothing.
-#   * --apply performs the REVERSIBLE local teardown:
-#       - stop + disable + remove the scrum-master's systemd --user units
-#       - `git submodule deinit` + `git rm` the scrum-master role dir + clean
-#         .gitmodules / .git/modules  (STAGED only — never commits or pushes)
-#       - remove the ~/.hermes/profiles/<sm-id> symlink
-#       - drop the scrum-master entry from the fleet registry (.bak first)
-#   * The IRREVERSIBLE `gh repo delete` of each scrum-master runtime repo is held
-#     behind BOTH --decommission-sm-remotes AND --yes-i-mean-it, and runs only
-#     after all local teardown has succeeded.
-#
-# It never commits or pushes any project repo — staged submodule removals are
-# left for you to review and commit (some live in sensitive repos: ~/docker,
-# ~/.config/zshyzsh).
-#
-# Usage:
-#   migrate-unify.sh                      # dry-run report (all scrum-masters)
-#   migrate-unify.sh --apply              # reversible local teardown
-#   migrate-unify.sh --apply --decommission-sm-remotes --yes-i-mean-it
-#                                         # + delete the runtime repos on GitHub
-#   migrate-unify.sh --agent pjangler-scrum-master [--apply ...]   # one only
-
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-
-HERMES_TEMPLATE_CONFIG="${HERMES_TEMPLATE_CONFIG:-${XDG_CONFIG_HOME:-$HOME/.config}/hermes-agent-template/config.toml}"
-cfg() {  # cfg <dotted.key> <default>
-  local key="$1" def="$2"
-  { [[ -f "$HERMES_TEMPLATE_CONFIG" ]] && command -v python3 >/dev/null 2>&1; } || { printf '%s' "$def"; return 0; }
-  python3 - "$HERMES_TEMPLATE_CONFIG" "$key" "$def" <<'PYEOF' || printf '%s' "$def"
-import sys, os
-try:
-    import tomllib
-except ModuleNotFoundError:
-    print(sys.argv[3], end=""); sys.exit(0)
-try:
-    cur = tomllib.load(open(sys.argv[1], "rb"))
-except Exception:
-    print(sys.argv[3], end=""); sys.exit(0)
-for p in sys.argv[2].split("."):
-    if isinstance(cur, dict) and p in cur:
-        cur = cur[p]
-    else:
-        print(sys.argv[3], end=""); sys.exit(0)
-print(os.path.expanduser(str(cur)), end="")
-PYEOF
-  return 0
-}
-
-FLEET_ENV="${HERMES_FLEET_ENV:-$(cfg fleet.fleet_env "$HOME/.hermes/fleet.env")}"
-[[ -f "$FLEET_ENV" ]] && { # shellcheck disable=SC1090
-  source "$FLEET_ENV"; }
-REGISTRY_FILE="${HERMES_FLEET_REGISTRY_FILE:-$(cfg fleet.registry_file "$HOME/.hermes/agents-registry.yaml")}"
-PROFILES_DIR="${HERMES_FLEET_HOME:-$HOME/.hermes}/profiles"
-SYS_DIR="$HOME/.config/systemd/user"
-
-APPLY=0
-DELETE_REMOTES=0
-CONFIRM=0
-declare -a ONLY_AGENTS=()
-while [[ $# -gt 0 ]]; do
-  case "$1" in
-    --apply) APPLY=1 ;;
-    --decommission-sm-remotes) DELETE_REMOTES=1 ;;
-    --yes-i-mean-it) CONFIRM=1 ;;
-    --all) : ;;  # accepted for symmetry; default is already all scrum-masters
-    --agent) shift; ONLY_AGENTS+=("$1") ;;
-    -h|--help) sed -n '3,55p' "$0"; exit 0 ;;
-    *) echo "migrate-unify: unknown flag $1" >&2; exit 2 ;;
-  esac
-  shift
-done
-
-[[ -f "$REGISTRY_FILE" ]] || { echo "migrate-unify: registry not found: $REGISTRY_FILE" >&2; exit 2; }
-command -v python3 >/dev/null || { echo "migrate-unify: python3 required" >&2; exit 2; }
-
-MODE="DRY-RUN"; [[ $APPLY -eq 1 ]] && MODE="APPLY"
-echo "migrate-unify — mode: $MODE   registry: $REGISTRY_FILE"
-echo "================================================================"
-
-# Emit scrum-master rows (pjangler first), TSV: agent_id, repo, role_dir, project_path, runtime_repo, profile_name
-read_sm() {
-  python3 - "$REGISTRY_FILE" <<'PYEOF'
-import sys, yaml
-data = yaml.safe_load(open(sys.argv[1], encoding="utf-8")) or {}
-rows = []
-for aid, a in (data.get("agents") or {}).items():
-    if not isinstance(a, dict) or a.get("role") != "scrum-master":
-        continue
-    rows.append((aid, str(a.get("repo") or ""), str(a.get("role_dir") or ""),
-                 str(a.get("project_path") or ""), str(a.get("runtime_repo") or ""),
-                 str(a.get("profile_name") or aid)))
-# pjangler first (the reference repo / canary), then alphabetical.
-rows.sort(key=lambda r: (r[1] != "pjangler", r[0]))
-for r in rows:
-    print("\t".join(r))
-PYEOF
-}
-
-wanted() {
-  [[ ${#ONLY_AGENTS[@]} -eq 0 ]] && return 0
-  local a; for a in "${ONLY_AGENTS[@]}"; do [[ "$a" == "$1" ]] && return 0; done; return 1
-}
-
-run() {  # run <human-description> -- <cmd...>   (prints always; executes only on --apply)
-  local desc="$1"; shift
-  [[ "${1:-}" == "--" ]] && shift
-  if [[ $APPLY -eq 1 ]]; then
-    printf '    [do]   %s\n' "$desc"
-    local rc=0
-    "$@" >/dev/null 2>&1 || rc=$?
-    if [[ $rc -ne 0 ]]; then printf '      (non-fatal: returned %s)\n' "$rc"; fi
-  else
-    printf '    [dry]  %s\n' "$desc"
-  fi
-}
-
-declare -a REMOVE_IDS=()       # registry ids to drop (batched)
-declare -a DELETE_REPOS=()     # runtime repos to delete (gated, last)
-declare -a STAGED_REPOS=()     # project repos left with staged submodule removals
-PROCESSED=0
-
-while IFS=$'\t' read -r sm_id repo role_dir project_path runtime_repo profile_name; do
-  wanted "$sm_id" || continue
-  PROCESSED=$((PROCESSED + 1))
-  echo
-  echo ">>> $sm_id   (repo=$repo)"
-  echo "    role_dir     : $role_dir"
-  echo "    project_path : $project_path"
-  echo "    runtime_repo : $runtime_repo"
-
-  # 1. systemd --user units: stop + disable + remove every hermes-<sm_id>-* unit.
-  shopt -s nullglob
-  units=("$SYS_DIR/hermes-${sm_id}-"*)
-  shopt -u nullglob
-  if [[ ${#units[@]} -eq 0 ]]; then
-    echo "    units        : (none found on disk)"
-  else
-    for uf in "${units[@]}"; do
-      u="$(basename "$uf")"
-      run "stop+disable $u" -- systemctl --user disable --now "$u"
-      run "rm unit file $u" -- rm -f "$uf"
-    done
-    if [[ $APPLY -eq 1 ]]; then
-      systemctl --user daemon-reload 2>/dev/null || true
-      systemctl --user reset-failed 2>/dev/null || true
-    fi
-  fi
-
-  # 2. scrum-master runtime submodule + role dir (STAGED removal, never committed).
-  if [[ -n "$project_path" && ( -d "$project_path/.git" || -f "$project_path/.git" ) ]]; then
-    sm_rel="agents/hermes/scrum-master"
-    if [[ -e "$project_path/$sm_rel" ]]; then
-      run "git submodule deinit -f $sm_rel/runtime" -- git -C "$project_path" submodule deinit -f "$sm_rel/runtime"
-      run "git rm -rf $sm_rel (staged)" -- git -C "$project_path" rm -rf "$sm_rel"
-      run "rm .git/modules/$sm_rel residue" -- rm -rf "$project_path/.git/modules/$sm_rel"
-      run "scrub .gitmodules section" -- git -C "$project_path" config -f "$project_path/.gitmodules" --remove-section "submodule.$sm_rel/runtime"
-      if [[ $APPLY -eq 1 ]]; then STAGED_REPOS+=("$project_path"); fi
-    else
-      echo "    role dir     : already absent at $project_path/$sm_rel"
-    fi
-  else
-    echo "    submodule    : project_path is not a git repo, skipping submodule teardown: $project_path"
-  fi
-
-  # 3. profile symlink ~/.hermes/profiles/<sm_id>.
-  link="$PROFILES_DIR/$profile_name"
-  if [[ -L "$link" ]]; then
-    run "rm profile symlink $link" -- rm -f "$link"
-  elif [[ -e "$link" ]]; then
-    echo "    profile      : $link is a real dir, NOT a symlink — left untouched (MANUAL)"
-  else
-    echo "    profile      : no symlink at $link"
-  fi
-
-  # 4. registry entry (batched; rewritten once at the end with a .bak).
-  REMOVE_IDS+=("$sm_id")
-  echo "    registry     : will drop agents.$sm_id"
-
-  # 5. runtime repo delete (gated, deferred to the very end).
-  [[ -n "$runtime_repo" ]] && DELETE_REPOS+=("$runtime_repo")
-done < <(read_sm)
-
-if [[ $PROCESSED -eq 0 ]]; then
-  echo; echo "migrate-unify: no scrum-master agents matched. Nothing to do."
-  exit 0
-fi
-
-# --- Registry rewrite (batched) ---
-echo
-if [[ ${#REMOVE_IDS[@]} -gt 0 ]]; then
-  if [[ $APPLY -eq 1 ]]; then
-    python3 - "$REGISTRY_FILE" "${REMOVE_IDS[@]}" <<'PYEOF'
-import sys, shutil, datetime, yaml
-path = sys.argv[1]; ids = sys.argv[2:]
-shutil.copy(path, path + ".bak-" + datetime.datetime.now().strftime("%Y%m%d%H%M%S"))
-data = yaml.safe_load(open(path, encoding="utf-8")) or {}
-agents = data.get("agents") or {}
-removed = [i for i in ids if agents.pop(i, None) is not None]
-data["agents"] = agents
-with open(path, "w", encoding="utf-8") as f:
-    f.write(yaml.safe_dump(data, sort_keys=False))
-print("    registry: removed " + ", ".join(removed) + " (backup written)")
-PYEOF
-  else
-    echo "    [dry]  would drop registry entries: ${REMOVE_IDS[*]} (after .bak)"
-  fi
-fi
-
-# --- Irreversible remote deletes (double-gated, last) ---
-echo
-echo "Runtime repos to delete on GitHub:"
-printf '    %s\n' "${DELETE_REPOS[@]}"
-if [[ ${#DELETE_REPOS[@]} -gt 0 ]]; then
-  if [[ $APPLY -eq 1 && $DELETE_REMOTES -eq 1 && $CONFIRM -eq 1 ]]; then
-    if command -v gh >/dev/null 2>&1; then
-      for r in "${DELETE_REPOS[@]}"; do
-        printf '    [do]   gh repo delete %s --yes\n' "$r"
-        gh repo delete "$r" --yes 2>&1 | tail -1 || echo "      (delete failed or already gone: $r)"
-      done
-    else
-      echo "    gh not on PATH — cannot delete remotes."
-    fi
-  else
-    echo "    [HELD] add --apply --decommission-sm-remotes --yes-i-mean-it to delete these (IRREVERSIBLE)."
-  fi
-fi
-
-echo
-echo "================================================================"
-echo "migrate-unify: processed $PROCESSED scrum-master agent(s) in $MODE mode."
-if [[ $APPLY -eq 1 && ${#STAGED_REPOS[@]} -gt 0 ]]; then
-  echo
-  echo "Review + commit the staged scrum-master removals in these repos (NOT auto-committed):"
-  printf '%s\n' "${STAGED_REPOS[@]}" | sort -u | sed 's/^/    git -C /; s/$/ status/'
-fi

package/templates/hermes-agent/scripts/momo-unify-agent.py

@@ -1,224 +0,0 @@
-#!/usr/bin/env python3
-"""momo-unify-agent — apply the Momo/Hermes unification to a DEPLOYED agent.
-
-Per role dir (agents/hermes/<role>) it:
-  1. renders SOUL.md from momo/spec/momo-agent.spec.yaml × the agent's role.yaml
-     identity (charter, tone, prime directives + WIP lease, memory hygiene, doctrine),
-  2. honcho-neutralizes runtime/config.yaml (memory.provider='', disabled_toolsets
-     += memory) so memory is the shared Hindsight bank only,
-  3. installs the shared WIP=1 lease: copies .scripts/momo-wip-lock.py and wires
-     .scripts/heartbeat.sh (if present) to acquire/release it around the reconcile pass.
-
-Idempotent; dry-run by default (--apply mutates). This is the existing-fleet
-counterpart to the copier template (future agents get the same via provisioning).
-Needs PyYAML — run with ~/.hermes/hermes-agent/.venv/bin/python.
-
-Usage: momo-unify-agent.py --role-dir <agents/hermes/pm> [--apply] [--spec P]
-"""
-from __future__ import annotations
-import argparse, sys
-from pathlib import Path
-try:
-    import yaml
-except ImportError:
-    sys.exit("needs PyYAML — run with ~/.hermes/hermes-agent/.venv/bin/python")
-
-SPEC_DEFAULT = Path("~/code/33GOD/momo/spec/momo-agent.spec.yaml").expanduser()
-LOCK_SRC = Path("~/code/33GOD/momo/skill/scripts/momo-wip-lock.py").expanduser()
-
-SOUL_TMPL = """# {display_name}
-
-You are **{display_name}** — the Momo PM/EM orchestrator for the `{repo}`
-repository, running as its autonomous **Hermes carrier**. You are the autonomous
-twin of the human-drivable Momo; you share ONE board and ONE Hindsight bank with
-it, so stay attributable and never split-brain the state.
-
-<!-- Rendered from momo/spec/momo-agent.spec.yaml (role: {role}) × this repo's identity.
-     Regenerate via momo-unify-agent.py / the Hermes adapter, not by hand. -->
-
-## Identity
-
-| | |
-| --- | --- |
-| Agent ID | `{agent_id}` |
-| Repo | `{repo}` |
-| Role | `{role}` |
-| Telegram | {telegram} |
-| Purpose | {purpose} |
-
-## Scope
-
-You operate only within the working directory of `{repo}`. Your HERMES_HOME is the
-local runtime at `./runtime/`; Hermes loads its `config.yaml` directly.
-
-## Tone
-
-{tone}
-
-## Role-specific behavior
-
-{charter}
-
-Prime directives (non-negotiable):
-
-- **Never mutate code.** Every code change flows through a delegated worker.
-- **WIP = 1**, shared with the human-drivable Momo via the driver lease
-  (`.scripts/momo-wip-lock.py` → `runtime/wip-driver.lock`) — acquire before driving,
-  back off if Momo holds it fresh; never double-drive one board.
-- **Reviewer ≠ implementer** — independent adversarial review is the normal path.
-- **Evidence over status** — a board column is a claim; repo evidence is proof.
-- **Everything is an event** — record consequential calls as Bloodbank decision events.
-- **Anti-stall** — never park a pass on operator sign-off.
-
-Default execution: {default_execution}.
-
-## Memory hygiene
-
-Your durable memory is the shared **Hindsight bank `{repo}`** — one bank per PROJECT,
-shared with the human-drivable Momo twin. Honcho and the per-agent `runtime/memories/`
-store are **neutralized**: do not rely on `MEMORY.md`/`USER.md`. Retain with
-`hindsight memory retain {repo} "…" --context <cat>`; recall with
-`hindsight memory recall {repo} "…"`.
-
-## Doctrine
-
-Decide on the operator's behalf using **`~/code/33GOD/momo/PILLARS.md`** (canonical,
-priority-ordered). This soul **references** that file; it does not copy it. Cite the
-pillar(s) that drove a consequential call in its decision event.
-"""
-
-LEASE_BLOCK = '''# Coexistence WIP=1 lease (momo E2/S2.3): don't full-drive if the human-drivable
-# Momo holds it. A crashed holder's lease expires (ttl) so the board is never wedged.
-WIP_LOCK="$RUNTIME/wip-driver.lock"
-if ! python3 "$ROLE_DIR/.scripts/momo-wip-lock.py" acquire "$WIP_LOCK" "hermes:$AGENT_ID" --ttl 3600 >/dev/null 2>&1; then
-  printf '[heartbeat] WIP lease held by Momo — skipping full reconcile pass this tick\\n'
-  maybe_checkpoint
-  exit 0
-fi
-trap 'python3 "$ROLE_DIR/.scripts/momo-wip-lock.py" release "$WIP_LOCK" "hermes:$AGENT_ID" >/dev/null 2>&1 || true' EXIT
-
-'''
-
-SPAWN_ANCHOR = 'prompt="$(<"$PROMPT_FILE")"'
-
-
-def load_role(role_dir: Path):
-    """role.yaml at the role dir, else inside runtime/."""
-    for p in (role_dir / "role.yaml", role_dir / "runtime" / "role.yaml"):
-        if p.is_file():
-            return yaml.safe_load(p.read_text()) or {}
-    return None
-
-
-def identity(r: dict, spec: dict) -> dict:
-    repo = r.get("repo", "?"); role = r.get("role", "pm")
-    agent_id = r.get("agent_id", f"{repo}-{role}")
-    disp = r.get("display_name") or f"{repo.title()} {role.upper() if len(role) <= 3 else role.title()}"
-    tel = (r.get("telegram") or {}).get("bot_username", "") if isinstance(r.get("telegram"), dict) else ""
-    tone_key = r.get("soul_tone") or spec["personality"]["default_tone"]
-    rs = spec["roles"].get(role, spec["roles"]["pm"])
-    return {
-        "repo": repo, "role": role, "agent_id": agent_id, "display_name": disp,
-        "telegram": f"@{tel}" if tel else "—",
-        "purpose": rs["purpose_template"].format(repo=repo),
-        "tone": spec["personality"]["tones"].get(tone_key, spec["personality"]["tones"]["direct"]),
-        "charter": rs.get("charter", "").strip(),
-        "default_execution": rs.get("default_execution", "subagent-driven-development (WIP=1, spec + quality gates)"),
-    }
-
-
-def neutralize_config(cfg: Path, apply: bool) -> str:
-    if not cfg.exists():
-        return "no config.yaml (skip honcho)"
-    base = yaml.safe_load(cfg.read_text()) or {}
-    already = (isinstance(base.get("memory"), dict) and base["memory"].get("provider") == ""
-               and "memory" in (base.get("agent", {}) or {}).get("disabled_toolsets", []))
-    if already:
-        return "honcho: already neutralized"
-    overlay = {"memory": {"provider": "", "memory_enabled": False, "user_profile_enabled": False},
-               "agent": {"disabled_toolsets": ["memory"]}}
-
-    def merge(b, o):
-        for k, v in o.items():
-            if isinstance(b.get(k), dict) and isinstance(v, dict):
-                merge(b[k], v)
-            else:
-                b[k] = v
-    merge(base, overlay)
-    if apply:
-        cfg.with_suffix(cfg.suffix + ".pre-unify.bak").write_text(cfg.read_text()) if not cfg.with_suffix(cfg.suffix + ".pre-unify.bak").exists() else None
-        yaml.safe_dump(base, cfg.open("w"), sort_keys=False)
-    return "honcho: neutralized"
-
-
-def install_lease(role_dir: Path, apply: bool) -> str:
-    scripts = role_dir / ".scripts"
-    msgs = []
-    if not scripts.is_dir():
-        return "no .scripts (skip lease)"
-    dst = scripts / "momo-wip-lock.py"
-    if not dst.exists() or dst.read_bytes() != LOCK_SRC.read_bytes():
-        if apply:
-            dst.write_bytes(LOCK_SRC.read_bytes()); dst.chmod(0o755)
-        msgs.append("lock helper installed")
-    else:
-        msgs.append("lock helper present")
-    hb = scripts / "heartbeat.sh"
-    if not hb.is_file():
-        msgs.append("no heartbeat.sh (reactive agent — lease unused)")
-        return "; ".join(msgs)
-    txt = hb.read_text()
-    if "momo-wip-lock.py" in txt:
-        msgs.append("heartbeat already lease-wired")
-    elif SPAWN_ANCHOR in txt:
-        if apply:
-            hb.write_text(txt.replace(SPAWN_ANCHOR, LEASE_BLOCK + SPAWN_ANCHOR, 1))
-        msgs.append("heartbeat lease-wired")
-    else:
-        msgs.append("WARN: heartbeat spawn anchor not found — wire manually")
-    return "; ".join(msgs)
-
-
-def unify(role_dir: Path, spec: dict, apply: bool, repo_override: str | None = None) -> None:
-    r = load_role(role_dir)
-    if r is None:
-        if not repo_override:
-            print(f"  SKIP: no role.yaml at {role_dir} (bare-layout agent — pass --repo to unify it)"); return
-        r = {"repo": repo_override, "role": "pm", "agent_id": f"{repo_override}-pm"}
-        print(f"  (synthesized identity for bare agent: {r['agent_id']})")
-    idt = identity(r, spec)
-    render = SOUL_TMPL.format(**idt)
-    # Write BOTH the role-dir source AND the LIVE runtime/SOUL.md that hermes
-    # actually loads (HERMES_HOME=runtime). Provisioning copies source->runtime,
-    # but for an already-deployed agent we must update the live copy directly.
-    for label, sp in (("src ", role_dir / "SOUL.md"),
-                      ("live", role_dir / "runtime" / "SOUL.md")):
-        if not sp.parent.exists():
-            continue
-        cur = sp.read_text() if sp.is_file() else ""
-        if cur.strip() == render.strip():
-            print(f"  SOUL[{label}]: already unified ({idt['agent_id']})")
-        else:
-            if apply:
-                sp.write_text(render)
-            print(f"  SOUL[{label}]: {'written' if apply else 'would write'} ({idt['agent_id']})")
-    print("  " + neutralize_config(role_dir / "runtime" / "config.yaml", apply))
-    print("  " + install_lease(role_dir, apply))
-
-
-def main() -> int:
-    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
-    ap.add_argument("--role-dir", required=True)
-    ap.add_argument("--spec", default=str(SPEC_DEFAULT))
-    ap.add_argument("--repo", default=None, help="synthesize identity for a bare agent with no role.yaml")
-    ap.add_argument("--apply", action="store_true")
-    a = ap.parse_args()
-    spec = yaml.safe_load(Path(a.spec).read_text())
-    role_dir = Path(a.role_dir).expanduser().resolve()
-    print(f"== unify ({'APPLY' if a.apply else 'dry-run'}) :: {role_dir}")
-    unify(role_dir, spec, a.apply, a.repo)
-    return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())

package/templates/hermes-agent/scripts/repair-runtime-checkpoint.sh

@@ -1,10 +0,0 @@
-#!/usr/bin/env bash
-# Retained only so older automation fails with a clear compatibility message.
-# The former checkpoint repair belongs to a retired persistence design.
-set -euo pipefail
-
-printf '%s\n' \
-  'ERROR: repair-runtime-checkpoint.sh is retired and intentionally non-operational.' \
-  'Current Hermes runtimes are ignored local state; this command changed nothing.' \
-  'See docs/operations.md for backup, restore, and retention policy.' >&2
-exit 64

package/templates/hermes-agent/scripts/retrofit-70-envfile.sh

@@ -1,129 +0,0 @@
-#!/usr/bin/env bash
-# Retrofit existing fleet systemd unit files to load runtime/.env.
-#
-# Companion to the fix in commit 3b7d2e5 (template/.scripts/70-systemd.sh).
-# Future agents pick the fix up automatically via pjangler; existing agents
-# already have unit files installed at ~/.config/systemd/user/ that won't
-# change until we rewrite them in place.
-#
-# What this does:
-#   - Finds every hermes-<agent>-{gateway,consumer}.service unit
-#   - Inserts `EnvironmentFile=-<runtime>/.env` after `Environment=HERMES_HOME=...`
-#     if not already present
-#   - daemon-reloads once
-#   - restarts each unit it modified (skippable with --no-restart)
-#
-# Idempotent: re-running is a no-op if the directive is already present.
-# Dry-run by default; pass --apply to actually write/restart.
-#
-# Usage:
-#   scripts/retrofit-70-envfile.sh             # dry-run: show what would change
-#   scripts/retrofit-70-envfile.sh --apply     # write + daemon-reload + restart
-#   scripts/retrofit-70-envfile.sh --apply --no-restart
-#
-set -euo pipefail
-
-APPLY=0
-RESTART=1
-for arg in "$@"; do
-  case "$arg" in
-    --apply)      APPLY=1 ;;
-    --no-restart) RESTART=0 ;;
-    -h|--help)
-      sed -n '2,/^set -euo/p' "$0" | sed 's/^# \{0,1\}//;/^set -euo/d'
-      exit 0
-      ;;
-    *) echo "unknown arg: $arg" >&2; exit 2 ;;
-  esac
-done
-
-SYS_DIR="$HOME/.config/systemd/user"
-DIRECTIVE_PREFIX="EnvironmentFile=-"
-
-shopt -s nullglob
-UNITS=( "$SYS_DIR"/hermes-*-gateway.service "$SYS_DIR"/hermes-*-consumer.service )
-shopt -u nullglob
-
-if [[ ${#UNITS[@]} -eq 0 ]]; then
-  echo "no hermes-*-{gateway,consumer}.service units found under $SYS_DIR"
-  exit 0
-fi
-
-mode="DRY-RUN"; [[ "$APPLY" == "1" ]] && mode="APPLY"
-echo "[$mode] scanning ${#UNITS[@]} unit(s) under $SYS_DIR"
-
-MODIFIED=()
-SKIPPED_OK=()
-SKIPPED_BAD=()
-
-for unit in "${UNITS[@]}"; do
-  base="$(basename "$unit")"
-
-  # Parse the runtime path from the Environment=HERMES_HOME= line.
-  runtime="$(grep -E '^Environment=HERMES_HOME=' "$unit" | head -1 | cut -d= -f3-)"
-  if [[ -z "$runtime" ]]; then
-    echo "  SKIP  $base — no Environment=HERMES_HOME= line; not a template-generated unit"
-    SKIPPED_BAD+=( "$base" )
-    continue
-  fi
-
-  expected="${DIRECTIVE_PREFIX}${runtime}/.env"
-
-  if grep -qxF "$expected" "$unit"; then
-    echo "  OK    $base — already has $expected"
-    SKIPPED_OK+=( "$base" )
-    continue
-  fi
-
-  echo "  PATCH $base"
-  echo "          + $expected"
-
-  if [[ "$APPLY" == "1" ]]; then
-    # Insert the directive on the line AFTER the Environment=HERMES_HOME= line.
-    # Use a tmp file + atomic replace to avoid half-written units.
-    tmp="$(mktemp "$unit.XXXXXX")"
-    awk -v ins="$expected" '
-      { print }
-      /^Environment=HERMES_HOME=/ && !done { print ins; done=1 }
-    ' "$unit" > "$tmp"
-    chmod --reference="$unit" "$tmp"
-    mv "$tmp" "$unit"
-  fi
-  MODIFIED+=( "$base" )
-done
-
-echo
-echo "summary:"
-echo "  ${#MODIFIED[@]} patched, ${#SKIPPED_OK[@]} already correct, ${#SKIPPED_BAD[@]} skipped (non-template)"
-
-if [[ "$APPLY" != "1" ]]; then
-  echo
-  echo "dry-run only. re-run with --apply to write the changes."
-  exit 0
-fi
-
-if [[ ${#MODIFIED[@]} -eq 0 ]]; then
-  echo "nothing to reload."
-  exit 0
-fi
-
-echo
-echo "systemctl --user daemon-reload"
-systemctl --user daemon-reload
-
-if [[ "$RESTART" != "1" ]]; then
-  echo "skipping restarts (--no-restart). reload only."
-  exit 0
-fi
-
-echo
-echo "restarting ${#MODIFIED[@]} modified unit(s)..."
-for unit in "${MODIFIED[@]}"; do
-  echo "  systemctl --user restart $unit"
-  if ! systemctl --user restart "$unit"; then
-    echo "    WARN: restart failed for $unit (status below)" >&2
-    systemctl --user --no-pager status "$unit" | head -10 >&2 || true
-  fi
-done
-
-echo "done."