Manager Setup

Manager

Manager mode has TWO independent layers. Keep them straight:

1. SOFT codewords (++m / ++a / ++rr / ++r) — autonomous, hook-driven, ALWAYS fire. A UserPromptSubmit hook (hooks/manager-prompt.mjs) watches every prompt; when it sees a codeword it injects the matching block as additionalContext for that one turn. This is NOT enabled/disabled by this skill — it works regardless of skill state. The skill only explains it (status) and customizes its TEXT (edit/purge). Detection (longest-prefix first within the review group):
   • ++m → Manager mode. PLAN-AWARE: when the session is in plan mode (permission_mode === 'plan') it injects the planmode block (full + plan addon — writes the task graph, uses the tasks tool); otherwise the plain full delegate-everything block. There is NO separate ++mp codeword.
   • ++a → Architecture-first directive (architect). Injects [DIRECTIVE: ARCHITECTURE-FIRST] before implementation — delegate an architecture pass that fits the project's existing architecture, patterns and rules; robust, scalable, and SIMPLE (no over-engineering); find the closest well-built counterpart in the repo and take its principles (additive to conventions/rules, not a replacement), clean seams. Independent group — combines with ++m and the review group. Mode-agnostic: same block in plan and normal mode (in plan mode it is written into the plan).
   • ++rr → Regression Review discipline (review-regression) — after each significant phase: no regression + project standard + correctness; two-phase review→double-check→fix; final cross-review at task end. Tested before ++r.
   • ++r → Review discipline (review-double) — two-phase multi-agent review→double-check→fix after each significant change; codeword-only (no ambient/wall injection).
   • When the HARD wall is ON, the Manager (full) block is ALSO auto-injected on EVERY turn — no codeword needed. Codewords and wall injection are independent.
2. HARD wall — opt-in, this skill only, PER-PROJECT, INSTALLED-INTO-THE-PROJECT, persistent. The wall is NOT a plugin hook. install does two things: it installs a self-contained PreToolUse guard into THIS project (copies the guard file + idempotently registers it in <cwd>/.claude/settings.local.json) and, only after the user has explicitly confirmed arming (P1 arm-confirmation gate — an AskUserQuestion answered "Yes, arm it now", or explicit wording like "enable the hard wall"/"включи хард уолл" already in the user's own prompt; the bare verb install/установи and any autonomy phrasing like "decide everything yourself"/"автономно" NEVER count as confirmation), arms it by flipping state.hard=true. Declining still installs the guard, leaving state.hard=false. The registered guard then physically denies mutating tools (Write/Edit/Bash/WebFetch/...) in the main session, leaving only delegate/read/track. Subagents stay fully free (agent_id linchpin). enable re-arms an already-installed wall — same confirmation gate, decline aborts with nothing changed; disable only flips state.hard=false — registration stays, the guard no-ops. uninstall removes the registration and the copied guard; purge also deletes the state file and the prompt overrides. The wall lives in project state + project settings, defaults OFF, persists until disable/uninstall. There is no codeword for the wall.

The two layers are orthogonal: the wall enforces delegation by removing hands; the codewords/prompt-text shape the Manager mindset. Either can be used alone.

INSTALL-ONCE + STATE-GATE (the safety crux): the guard is registered once in settings.local.json (a personal, gitignored file) but is gated at runtime by project state.json {hard}. Registration is the persistent plumbing; state.hard is the live kill-switch. This split exists because while the wall is armed it DENIES Edit/Bash on arbitrary files — so disable must NOT touch settings.local.json (that edit would be blocked). Instead disable flips state.json with the ONE Bash shape the guard self-exempts, so the state flip always succeeds even at level strict. Conclusion: state.json is the runtime kill-switch; registration is harmless inert plumbing left in place.

THE EXEMPT COMMAND (memorize this shape — nothing else gets through an armed wall):

node <ABS project root>/.claude/brewtools/manager/manager-state.mjs set hard=false

The guard exempts it only when ALL of these hold: the command starts with node , the FIRST argument after node resolves (realpath) to the helper this project actually installed — <root>/.claude/brewtools/manager/manager-state.mjs, or the plugin's own hooks/lib/manager-state.mjs next to the guard — there is no shell operator outside quotes, no $ expansion, no eval flag (-e/--eval/-p/--print/--input-type/--require/--import/--loader), and the remaining arguments are the helper's own CLI (get | set hard=<true|false> level=<strict|balanced> mcpAllow=<mcp__srv__tool[,...]|> [--cwd DIR]). No BT_ROOT= prelude, no && echo, no || echo, no test -f — every one of those is a shell operator and turns the exemption OFF. A file merely named manager-state.mjs elsewhere on disk is NOT exempt: the anchor is the absolute installed path, not the filename or a path suffix. install/upgrade copy the helper into the project precisely so this command needs no path resolution.

What the guard actually enforces (read this before you need it)

Recovery, in order of preference.

1. node <ABS root>/.claude/brewtools/manager/manager-state.mjs set hard=false — works at every level, including when state.json is corrupt (it rewrites the file).
2. A genuinely read-only MCP tool denied by the classifier — allowlist it (balanced only), same self-exempt command shape, quote the value so the shell keeps *: node <ABS root>/.claude/brewtools/manager/manager-state.mjs set 'mcpAllow=mcp__semble_code__*,mcp__github__get_file' — the list is replaced wholesale, one invalid entry writes nothing (exit 2), and set 'mcpAllow=' clears it.
3. Delegate: Task is always allowed and subagents are unwalled, so a subagent can run /brewtools:manager-setup upgrade or repair state for you.
4. Two residual cases need action OUTSIDE the session, and there is no in-session workaround — do not go hunting for one:
   • Claude Code changes the PreToolUse payload shape so the guard cannot parse it. Every main-session mutation is then denied. Fix: quit and delete the brewtools-manager-guard entry from .claude/settings.local.json in an editor.
   • Deleting the whole .claude/brewtools/ tree disarms the wall (no manager directory = never installed). That is the documented consequence of a manual rm -rf, not a way to disable the wall — use disable or uninstall, which keep settings and files consistent.

Prompt contract

Position 1 of $ARGUMENTS is a free-form prompt (RU/EN) — modes and flags are optional and may follow in any order. Nobody types keys: resolve the action + scope FROM the prompt via references/intent-routing.md (P0 table below is the same routing, EN/RU keyword split).

1. Strip flags. An explicit action token anywhere wins outright, no scoring.
2. Else score actions by distinct whole-word keyword hits (P0 table). Highest unique score wins. Tie with a destructive action (purge) -> AskUserQuestion; tie with status -> status; tie of two mutating actions -> the keyword appearing first; all zero -> status.
3. Empty arguments -> status; ask ONE scoping AskUserQuestion only when the answer changes what gets written or armed. status asks nothing.
4. Outcome-changing ambiguity (incl. hard-one-shot vs manager-run, enable vs disable) -> ONE AskUserQuestion (max 4 questions) BEFORE any work — this is P1.
5. Prose that is not an action/id/path is still input: extract the task from <task> в хард режиме / <task> от роли менеджера rather than treating the first word as a positional id.

Then print this block ONCE, after P1 and before the first action (P2). A read-only status prints it immediately before its report:

PLAN — brewtools:manager-setup
INPUT:  <arguments verbatim, or "(empty)">
MODE:   <resolved action> — <explicit | matched keyword: X | default>
SCOPE:  <resolved paths / level / task> — LAYER: <codewords (soft, always-on, hook-driven) |
        HARD wall (opt-in, this project, PreToolUse guard)> — name which layer this action touches
DO:     <2-5 imperative bullets>
RESULT: <what the user ends up holding>

Labels are literal; values follow the conversation language. install/upgrade/enable/ disable/uninstall/purge/level touch the HARD-wall layer; edit touches the codewords layer (prompt text only); hard-one-shot touches BOTH (arms/disarms the wall AND runs the task under the codewords contract); manager-run/inline-run touch only the codewords layer.

Robustness Rules

Scope, said once so it is never confused

"Wall scope" is fixed (project). "Prompt-text override scope" is a different, independent axis that edit/purge may target globally. Do not let --scope global leak onto the wall — it has no meaning there.

BT_ROOT Resolver

The plugin root is resolved from the skill's OWN directory (the CLAUDE_SKILL_DIR prompt substitution), never from CLAUDE_PLUGIN_ROOT -- that env var is not exported to a skill's Bash tool. Every Bash block resolves BT_ROOT this way (no hardcoded version):

SD="${CLAUDE_SKILL_DIR}"
if [ -n "$SD" ] && [ -f "$SD/../../.claude-plugin/plugin.json" ]; then BT_ROOT=$(cd "$SD/../.." && pwd); else BT_ROOT=$(ls -d ~/.claude/plugins/cache/claude-brewcode/brewtools/*/ 2>/dev/null | sort -V | tail -1 | sed 's:/*$::'); fi
[ -n "$BT_ROOT" ] || { echo "ERROR: cannot locate brewtools plugin root -- install/update brewtools first."; exit 1; }
test -f "$BT_ROOT/hooks/lib/manager-state.mjs" || { echo "❌ BT_ROOT invalid: $BT_ROOT"; exit 1; }

Paths (use $BT_ROOT literally in Bash):

• State helper: $BT_ROOT/hooks/lib/manager-state.mjs — exports resolveState, writeState, resolveStatePath; also a CLI: node <path>/manager-state.mjs get|set hard=<true|false> level=<strict|balanced> mcpAllow=<mcp__srv__tool[,...]|> [--cwd DIR]
• Prompt helper: $BT_ROOT/hooks/lib/manager-prompts.mjs — exports resolvePrompt, resolvePromptPath
• Guard source (shipped, self-contained, NOT in plugin hooks.json): $BT_ROOT/hooks/hardmode-guard.mjs — install/upgrade copy this into the project
• Plugin default blocks: $BT_ROOT/skills/manager-setup/references/<mode>.md (full.md, planmode.md)
• Wall policy + canonical status text: $BT_ROOT/skills/manager-setup/references/hard.md — Read it for the install model, status explainer and the allowlist details.

Project install targets (resolved from process.cwd()):

• Copied guard: <cwd>/.claude/brewtools/manager/hardmode-guard.mjs
• Copied state helper: <cwd>/.claude/brewtools/manager/manager-state.mjs — the off-switch CLI. Copied so disable/level are a fixed path needing no BT_ROOT resolution (resolution needs shell operators, which the armed wall denies)
• Registration: <cwd>/.claude/settings.local.json — a PreToolUse matcher "*" entry whose command runs node <ABS path to copied guard>. Tagged with marker brewtools-manager-guard so uninstall/purge can find it.

Resolution chains (must match helpers exactly)

The wall flags (hard/level) are resolved PROJECT-ONLY in code — the global state.json does NOT enable the wall. The skill writes them to project scope only. (The informational mode field may still resolve from global; hard/level do not.)

P0: Resolve Intent

Parse $ARGUMENTS (or the user's NL prompt, RU+EN) into { action, scope, mode, level, task } using references/intent-routing.md — Read and follow it.

Actions, canonical order: status, install, upgrade, enable, disable, uninstall, purge, plus the extras level <strict|balanced>, edit, and the run actions hard-one-shot, manager-run, inline-run.

on / off / reset / setup / remove are REMOVED as command words. on and off survive only as free-text synonyms routed to enable / disable above; reset routes to purge. Never print them as commands.

Arming is never automatic. install, enable, and hard-one-shot are the only actions that can write state.hard=true. Resolving the action (even from an explicit keyword like install, or autonomy phrasing like "autonomous, decide everything yourself"/"автономно выбирай сам") is NOT confirmation to arm — that confirmation is a separate, mandatory P1 gate. See P1 below.

Prompt-text override scope (ONLY for edit/purge): default = project. --scope global OR глобально / globally → global. This scope does NOT apply to install/upgrade/enable/disable/uninstall/level (those are project-only).

P1: Echo + Disambiguate

Print ONE line stating the resolved intent, e.g.:

Understood: install + arm the hard wall (project), level=balanced

If the action is ambiguous or signals conflict (e.g. enable + disable, a task that might be hard-one-shot vs manager-run, control implied but no verb) → AskUserQuestion with the candidate actions as options. Otherwise proceed.

Distinguish carefully: hard-one-shot (task + "в хард режиме"/"in hard mode") flips the wall and auto-reverts; manager-run (task + "от роли менеджера"/"as manager") never touches the wall, discipline by prompt only. If both/neither marker is present and a task exists, ask.

Arm-confirmation gate (unconditional — not only on ambiguity). install, enable, and hard-one-shot are the only actions that can write state.hard=true. Before P2 runs any of them, check whether the user's OWN prompt already carries EXPLICIT confirming wording: the words "hard wall" / "хард уолл" / "стену" TOGETHER with an enable/arm verb ("enable"/"arm"/"включи"/"заarmи"), e.g. "enable the hard wall", "arm the wall", "включи Hard Wall", "включи хард уолл", "заarmи стену". The bare skill verb alone (install/установи/enable/включи with no "wall" wording) does not count, and no autonomy-permission phrasing ever counts ("autonomous, decide everything yourself", "автономно", "выбирай сам" — these NEVER satisfy the gate). If explicit wording is present, treat the arm as already confirmed and skip the question. Otherwise ask exactly ONE AskUserQuestion before P2:

• install: "Arm the HARD wall in this project now? It will block Write/Edit/Bash in the MAIN session (subagents stay free) until disabled." — options Yes, arm it now / No, just install (stay disarmed).
• enable / hard-one-shot: the same question, options Yes, arm it now / No, cancel.

Record the outcome as ARM (true/false) for P2:

• install proceeds either way — it always copies the guard and registers the hook; ARM only decides whether state.hard is written true or false. ARM=false → report "installed but NOT armed — run enable (or confirm next time) to arm it."
• enable / hard-one-shot: ARM=false aborts the whole action before running anything — do not touch state.hard, report plainly that nothing changed.

P2: Execute

Print the ## Prompt contract PLAN block first — INPUT/MODE from P0, SCOPE naming which layer (codewords vs HARD wall, or both for hard-one-shot) — before running the mapped section below. status prints the same block immediately before its report instead of before a mutation.

Sections below map 1:1 to action: install, upgrade, enable, disable, uninstall, purge, level, status, edit, and the three run actions.

install (INSTALL + ARM the HARD wall — project only)

install is a five-step sequence: (1) set arm state per the P1 confirm gate (ARM), (2) copy the guard into the project, (3) idempotently register it in settings.local.json, (4) turn the task-graph tools on in that same file, (5) report whether a /reload is needed. All five run in ONE node Bash block so the registration is atomic and self-contained. ARM (true/false) is resolved by the P1 arm-confirmation gate — substitute it into the block below before running, same as LEVEL in the level action. The block:

• sets state.hard = (ARM === 'true') via writeState('project', {hard:arm}) — arms only when P1 confirmed,
• copies $BT_ROOT/hooks/hardmode-guard.mjs → <cwd>/.claude/brewtools/manager/hardmode-guard.mjs and $BT_ROOT/hooks/lib/manager-state.mjs → <cwd>/.claude/brewtools/manager/manager-state.mjs (both overwritten on EVERY install, so plugin updates propagate; the second one is the off-switch CLI),
• read-merge-atomic-writes <cwd>/.claude/settings.local.json, adding a PreToolUse matcher "*" entry that runs node <ABS copied-guard path> tagged brewtools-manager-guard, but ONLY if no entry already points at the manager guard (idempotent — running twice = ONE entry),
• in that SAME merge sets env.CLAUDE_CODE_ENABLE_TODO_TOOLS = "1" when Claude Code is >= 2.1.233 — creating the env object if absent, preserving every other key. This is unconditional and never asks: from 2.1.233 TaskCreate/TaskUpdate/TaskGet/TaskList are gated OFF by default, and the manager framework has no task graph without them. Below 2.1.233 the var does nothing and the tools are on anyway, so the write is skipped and the block says so,
• prints newlyRegistered and todoTools so you know whether to surface the /reload note and what happened to the task tools.

EXECUTE using Bash tool:

SD="${CLAUDE_SKILL_DIR}"
if [ -n "$SD" ] && [ -f "$SD/../../.claude-plugin/plugin.json" ]; then BT_ROOT=$(cd "$SD/../.." && pwd); else BT_ROOT=$(ls -d ~/.claude/plugins/cache/claude-brewcode/brewtools/*/ 2>/dev/null | sort -V | tail -1 | sed 's:/*$::'); fi
[ -n "$BT_ROOT" ] || { echo "ERROR: cannot locate brewtools plugin root -- install/update brewtools first."; exit 1; }
test -f "$BT_ROOT/hooks/hardmode-guard.mjs" || { echo "❌ BT_ROOT invalid: $BT_ROOT"; exit 1; }
ROOT=$(if [ -n "$CLAUDE_PROJECT_DIR" ] && [ -d "$CLAUDE_PROJECT_DIR" ]; then printf %s "$CLAUDE_PROJECT_DIR"; elif r=$(git rev-parse --show-toplevel 2>/dev/null) && [ -n "$r" ]; then printf %s "$r"; else d=$PWD; while [ "$d" != "/" ]; do if [ -d "$d/.git" ] || [ -d "$d/.claude" ]; then printf %s "$d"; break; fi; d=$(dirname "$d"); done; fi)
[ -n "$ROOT" ] || { echo "❌ cannot resolve project root — looked for CLAUDE_PROJECT_DIR, git toplevel, then .git/.claude above $PWD; nothing written"; exit 1; }
CCVER=$(claude --version 2>/dev/null | grep -oE '[0-9]+\.[0-9]+\.[0-9]+' | head -1 || true)
ARM=<true|false>
node --input-type=module -e "
import {writeState} from '${BT_ROOT}/hooks/lib/manager-state.mjs';
import fs from 'node:fs'; import path from 'node:path';
const cwd = '${ROOT}';
const arm = '${ARM}' === 'true';
const src = '${BT_ROOT}/hooks/hardmode-guard.mjs';
const dir = path.join(cwd, '.claude', 'brewtools', 'manager');
const guard = path.join(dir, 'hardmode-guard.mjs');
const helper = path.join(dir, 'manager-state.mjs');
const settings = path.join(cwd, '.claude', 'settings.local.json');
const TAG = 'brewtools-manager-guard';
// Task graph: CC 2.1.233+ gates TaskCreate/Update/Get/List off unless env.CLAUDE_CODE_ENABLE_TODO_TOOLS
// is set. Numeric compare, never string. Below 2.1.233 the key is a no-op, so skip the write.
const ccVer = '${CCVER}';
const geVersion = (v, t) => { const a = String(v).split('.').map(n => parseInt(n, 10)); return a.length === 3 && !a.some(Number.isNaN) && (a[0] - t[0] || a[1] - t[1] || a[2] - t[2]) >= 0; };
const todoToolsGated = geVersion(ccVer, [2,1,233]);
const todoTools = todoToolsGated ? 'enabled (CC ' + ccVer + ')'
  : ccVer ? 'skipped — CC ' + ccVer + ' predates the 2.1.233 gate, task tools are on by default'
  : 'skipped — could not read the Claude Code version; on 2.1.233+ set env.CLAUDE_CODE_ENABLE_TODO_TOOLS=1 by hand';
// 1. arm only if P1 confirmed
await writeState('project', {hard:arm}, cwd);
// 2. copy guard + off-switch CLI (overwrite each install)
fs.mkdirSync(dir, {recursive:true});
fs.copyFileSync(src, guard);
fs.copyFileSync('${BT_ROOT}/hooks/lib/manager-state.mjs', helper);
// 3. idempotent register, under the settings lock
const lock = settings + '.lock';
fs.mkdirSync(path.dirname(settings), {recursive:true});
let held = false;
for (let i = 0; i < 50 && !held; i++) {
  try { fs.mkdirSync(lock); held = true; }
  catch {
    try { if (Date.now() - fs.statSync(lock).mtimeMs > 30000) { fs.rmSync(lock, {recursive:true, force:true}); continue; } } catch {}
    Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, 100);
  }
}
if (!held) { console.error('ABORT: ' + lock + ' is held by another setup skill — retry in a moment; nothing was written'); process.exit(1); }
let newlyRegistered = false;
try {
  let cfg = {};
  try { cfg = JSON.parse(fs.readFileSync(settings,'utf8')); }
  catch (e) { if (e.code !== 'ENOENT') { console.error('ABORT: ' + settings + ' unreadable or invalid JSON (' + e.message + ') — fix it by hand; nothing was written'); process.exit(1); } }
  if (!cfg || typeof cfg !== 'object' || Array.isArray(cfg)) { console.error('ABORT: ' + settings + ' is not a JSON object — fix it by hand; nothing was written'); process.exit(1); }
  cfg.hooks = (cfg.hooks && typeof cfg.hooks==='object') ? cfg.hooks : {};
  const arr = Array.isArray(cfg.hooks.PreToolUse) ? cfg.hooks.PreToolUse : [];
  const has = m => Array.isArray(m.hooks) && m.hooks.some(h => typeof h.command==='string' && (h.command.includes(TAG) || h.command.includes('hardmode-guard.mjs')));
  if (!arr.some(has)) {
    arr.push({ matcher:'*', hooks:[{ type:'command', command:\`node \"\${guard}\" # \${TAG}\`, timeout:5 }] });
    newlyRegistered = true;
  }
  cfg.hooks.PreToolUse = arr;
  // 4. task graph on, same merge — idempotent, one key, every other setting preserved.
  if (todoToolsGated) { cfg.env = (cfg.env && typeof cfg.env==='object' && !Array.isArray(cfg.env)) ? cfg.env : {}; cfg.env.CLAUDE_CODE_ENABLE_TODO_TOOLS = '1'; }
  if (fs.existsSync(settings)) fs.copyFileSync(settings, settings + '.bak');
  const tmp = settings + '.tmp';
  fs.writeFileSync(tmp, JSON.stringify(cfg, null, 2) + '\n', 'utf8');
  fs.renameSync(tmp, settings);
} finally { fs.rmSync(lock, {recursive:true, force:true}); }
console.log(JSON.stringify({armed:arm, guard, helper, settings, newlyRegistered, todoTools, root:cwd}));
" && echo "✅ wall install step done (see armed field above)" || echo "❌ FAILED install wall"

Root, lock, backup — the three invariants every settings-writing block here shares. ROOT is the canonical recipe (CLAUDE_PROJECT_DIR → git rev-parse --show-toplevel → upward walk for .git/.claude → abort). An installer that cannot name its root ABORTS non-zero and says what it looked for — it never writes to a guessed root, and it never uses raw $PWD, which drifts to whatever subdirectory the session wandered into. The settings.local.json.lock directory is an O_EXCL mutex (stale-broken after 30 s) so two setup skills running in parallel cannot lose each other's edit; the file is re-read INSIDE the lock. settings.local.json.bak is written before every rename. A read that is not ENOENT, or content that is not a JSON object, ABORTS before anything is staged — a malformed settings file is never "the file is empty".

After the block:

• If armed:true → tell the user the exit command verbatim, with the real absolute path: node <ABS project root>/.claude/brewtools/manager/manager-state.mjs set hard=false — or just /brewtools:manager-setup disable, which runs exactly that.
• If armed:false → tell the user plainly: installed but NOT armed — the guard is in place but state.hard=false; run /brewtools:manager-setup enable (or answer "Yes, arm it now" next time) to arm it.
• If newlyRegistered:true → tell the user verbatim: Hook installed in .claude/settings.local.json — run /reload (or restart the session) for the wall to take effect.
• If newlyRegistered:false → the entry already existed; a reload is only needed if armed:true just flipped a previously-disarmed state.
• Report todoTools in one line: enabled → say TaskCreate/TaskUpdate/TaskGet/TaskList enabled via env.CLAUDE_CODE_ENABLE_TODO_TOOLS in .claude/settings.local.json; skipped → print the reason verbatim and move on.

The command in the registered entry uses an ABSOLUTE path to the copied guard and a # brewtools-manager-guard tag comment so uninstall can find it. Scope is always project — there is no global wall, never pass 'global'.

upgrade (re-emit the guard from the current plugin version — arm state kept, provenance restamped)

upgrade replays the install against the CURRENT plugin version so a claude plugin update finally reaches an already-installed project: it re-copies hardmode-guard.mjs and manager-state.mjs, re-registers the entry if it went missing, and — in the same read-merge-atomic-write of settings.local.json — sets env.CLAUDE_CODE_ENABLE_TODO_TOOLS = "1" on Claude Code >= 2.1.233. A project installed before the off-switch CLI existed has no project copy of manager-state.mjs, and one installed before the task-tool gate has no env key; upgrade is what backfills both, so run it once after updating brewtools. It asks nothing.

It restamps state.json, and ONLY the metadata trio. setup-status row 8 reads the top-level "version" of .claude/brewtools/manager/state.json as the headline; the guard's brewcode-meta: line is SECOND precedence, consulted only when that key is absent. So an upgrade that re-copied the guard but left state.json alone reported the old version forever and status printed stale after every upgrade — the staleness could never be cleared. The fix is the docsync-setup shape (brewdoc/skills/docsync-setup/SKILL.md mode upgrade): call writeState('project', {}, cwd) — an EMPTY partial. writeState merges {...existing, ...partial} and then stamps version / generated_by / last_updated, so with nothing in the partial it rewrites the trio and nothing else. hard and level are preserved byte-for-byte out of the existing file: a disarmed wall stays disarmed, an armed one stays armed, a customized level survives. That is what stateUntouched used to promise and it still holds for the ARM state — the block now reports armStatePreserved + stateRestamped so the two are not conflated.

It ABORTS when the project has no wall installed. upgrade must never be a back door that arms a wall the user never asked for — an uninstalled project is told to run install.

EXECUTE using Bash tool:

SD="${CLAUDE_SKILL_DIR}"
if [ -n "$SD" ] && [ -f "$SD/../../.claude-plugin/plugin.json" ]; then BT_ROOT=$(cd "$SD/../.." && pwd); else BT_ROOT=$(ls -d ~/.claude/plugins/cache/claude-brewcode/brewtools/*/ 2>/dev/null | sort -V | tail -1 | sed 's:/*$::'); fi
[ -n "$BT_ROOT" ] || { echo "ERROR: cannot locate brewtools plugin root -- install/update brewtools first."; exit 1; }
test -f "$BT_ROOT/hooks/hardmode-guard.mjs" || { echo "❌ BT_ROOT invalid: $BT_ROOT"; exit 1; }
ROOT=$(if [ -n "$CLAUDE_PROJECT_DIR" ] && [ -d "$CLAUDE_PROJECT_DIR" ]; then printf %s "$CLAUDE_PROJECT_DIR"; elif r=$(git rev-parse --show-toplevel 2>/dev/null) && [ -n "$r" ]; then printf %s "$r"; else d=$PWD; while [ "$d" != "/" ]; do if [ -d "$d/.git" ] || [ -d "$d/.claude" ]; then printf %s "$d"; break; fi; d=$(dirname "$d"); done; fi)
[ -n "$ROOT" ] || { echo "❌ cannot resolve project root — looked for CLAUDE_PROJECT_DIR, git toplevel, then .git/.claude above $PWD; nothing written"; exit 1; }
CCVER=$(claude --version 2>/dev/null | grep -oE '[0-9]+\.[0-9]+\.[0-9]+' | head -1 || true)
node --input-type=module -e "
import fs from 'node:fs'; import path from 'node:path';
import {writeState, resolveStatePath} from '${BT_ROOT}/hooks/lib/manager-state.mjs';
const cwd = '${ROOT}';
const src = '${BT_ROOT}/hooks/hardmode-guard.mjs';
const dir = path.join(cwd, '.claude', 'brewtools', 'manager');
const guard = path.join(dir, 'hardmode-guard.mjs');
const settings = path.join(cwd, '.claude', 'settings.local.json');
const TAG = 'brewtools-manager-guard';
const has = m => Array.isArray(m.hooks) && m.hooks.some(h => typeof h.command==='string' && (h.command.includes(TAG) || h.command.includes('hardmode-guard.mjs')));
// Task graph: CC 2.1.233+ gates TaskCreate/Update/Get/List off unless env.CLAUDE_CODE_ENABLE_TODO_TOOLS
// is set. Numeric compare, never string. Below 2.1.233 the key is a no-op, so skip the write.
const ccVer = '${CCVER}';
const geVersion = (v, t) => { const a = String(v).split('.').map(n => parseInt(n, 10)); return a.length === 3 && !a.some(Number.isNaN) && (a[0] - t[0] || a[1] - t[1] || a[2] - t[2]) >= 0; };
const todoToolsGated = geVersion(ccVer, [2,1,233]);
const todoTools = todoToolsGated ? 'enabled (CC ' + ccVer + ')'
  : ccVer ? 'skipped — CC ' + ccVer + ' predates the 2.1.233 gate, task tools are on by default'
  : 'skipped — could not read the Claude Code version; on 2.1.233+ set env.CLAUDE_CODE_ENABLE_TODO_TOOLS=1 by hand';
const lock = settings + '.lock';
fs.mkdirSync(path.dirname(settings), {recursive:true});
let held = false;
for (let i = 0; i < 50 && !held; i++) {
  try { fs.mkdirSync(lock); held = true; }
  catch {
    try { if (Date.now() - fs.statSync(lock).mtimeMs > 30000) { fs.rmSync(lock, {recursive:true, force:true}); continue; } } catch {}
    Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, 100);
  }
}
if (!held) { console.error('ABORT: ' + lock + ' is held by another setup skill — retry in a moment; nothing was written'); process.exit(1); }
let newlyRegistered = false;
try {
  let cfg = {};
  try { cfg = JSON.parse(fs.readFileSync(settings,'utf8')); }
  catch (e) { if (e.code !== 'ENOENT') { console.error('ABORT: ' + settings + ' unreadable or invalid JSON (' + e.message + ') — fix it by hand; nothing was written'); process.exit(1); } }
  if (!cfg || typeof cfg !== 'object' || Array.isArray(cfg)) { console.error('ABORT: ' + settings + ' is not a JSON object — fix it by hand; nothing was written'); process.exit(1); }
  const arr = (cfg.hooks && Array.isArray(cfg.hooks.PreToolUse)) ? cfg.hooks.PreToolUse : [];
  if (!arr.some(has) && !fs.existsSync(guard)) { console.error('ABORT: the wall is not installed in this project — run install instead'); process.exit(1); }
  fs.mkdirSync(dir, {recursive:true});
  fs.copyFileSync(src, guard);
  fs.copyFileSync('${BT_ROOT}/hooks/lib/manager-state.mjs', path.join(dir, 'manager-state.mjs'));
  if (!arr.some(has)) { arr.push({ matcher:'*', hooks:[{ type:'command', command:\`node \"\${guard}\" # \${TAG}\`, timeout:5 }] }); newlyRegistered = true; }
  cfg.hooks = (cfg.hooks && typeof cfg.hooks==='object') ? cfg.hooks : {};
  cfg.hooks.PreToolUse = arr;
  if (todoToolsGated) { cfg.env = (cfg.env && typeof cfg.env==='object' && !Array.isArray(cfg.env)) ? cfg.env : {}; cfg.env.CLAUDE_CODE_ENABLE_TODO_TOOLS = '1'; }
  if (fs.existsSync(settings)) fs.copyFileSync(settings, settings + '.bak');
  const tmp = settings + '.tmp';
  fs.writeFileSync(tmp, JSON.stringify(cfg, null, 2) + '\n', 'utf8');
  fs.renameSync(tmp, settings);
} finally { fs.rmSync(lock, {recursive:true, force:true}); }
// Restamp the metadata trio ONLY — empty partial, so hard/level/mode and every
// unknown key merge through from the existing file untouched.
let before = null;
try { before = JSON.parse(fs.readFileSync(resolveStatePath('project', cwd),'utf8')); } catch {}
const w = await writeState('project', {}, cwd);
const armStatePreserved = !before || (w.state.hard === before.hard && w.state.level === before.level);
console.log(JSON.stringify({guardReplaced:true, guard, newlyRegistered, todoTools,
  stateRestamped:{version:w.state.version, generated_by:w.state.generated_by, last_updated:w.state.last_updated},
  hard:w.state.hard, level:w.state.level, armStatePreserved}));
" && echo "✅ wall upgraded (arm state preserved, state.json restamped)" || echo "❌ FAILED upgrade"

Surface the /reload note only when newlyRegistered:true. Report todoTools in one line the same way install does — an old project that predates the key gets it backfilled here.

enable (ARM an installed wall — state flip only)

Gated by P1. This block only runs if the P1 arm-confirmation gate resolved ARM=true (explicit wording in the user's prompt, or an AskUserQuestion answered "Yes, arm it now"). If P1 resolved ARM=false, do NOT run this block at all — report plainly that arming was declined and nothing changed, and stop.

enable flips state.hard=true and nothing else. If the wall was never installed there is no guard to arm, so the block reports notInstalled instead of writing a state

…(truncated)