crewforth 3.0.0-rc.1 → 3.0.0-rc.3

package/kit/eval/smoke-test.sh

@@ -1575,6 +1575,25 @@ if [ "$IS_KIT" = 1 ]; then
   grep -q 'N_AG="$(count_installed' "$KR/start.sh" 2>/dev/null \
     && pass "start.sh derives its summary counts from the payload" \
     || fail "start.sh no longer derives its summary counts from the payload"
+  # ...and the skills/commands it prints are the README's, not the directory count: the RC-1 rehearsal printed
+  # "51 skills" beside a README that says 39 skills and 10 commands. count_showcase is lifted out of start.sh and
+  # driven on a three-entry fixture (a skill, a command, an experimental one) and on the real payload, whose answer
+  # must equal the counts the README gates above use.
+  _csf="$(sed -n '/^count_showcase() {/,/^}/p' "$KR/start.sh")"
+  if [ -z "$_csf" ]; then fail "start.sh has no count_showcase — the install prints a directory count again"
+  else
+    _csd="$(mktemp -d)"; mkdir -p "$_csd/a" "$_csd/b" "$_csd/c"
+    printf -- '---\nname: a\ndescription: x\n---\n' > "$_csd/a/SKILL.md"
+    printf -- '---\nname: b\ndescription: x\nmetadata:\n  kind: command\n---\n' > "$_csd/b/SKILL.md"
+    printf -- '---\nname: c\ndescription: x\nmetadata:\n  kind: command\n  experimental: true\n---\n' > "$_csd/c/SKILL.md"
+    _csx="$(bash -c "$_csf"'; count_showcase "$1"; echo "$N_SKL $N_CMD"' _ "$_csd")"
+    _cshown="$(bash -c "$_csf"'; count_showcase "$1"; echo "$N_SKL $N_CMD"' _ "$SKILLS")"
+    rm -rf "$_csd"
+    if [ "$_csx" != "1 1" ]; then fail "count_showcase miscounts its fixture: '$_csx', want '1 1' (the experimental entry counts as neither)"
+    elif _nsk=0; for _sf in "$SKILLS"/*/SKILL.md; do _sn="${_sf%/SKILL.md}"; _sn="${_sn##*/}"; is_cmd "$_sn" || is_exp "$_sn" || _nsk=$((_nsk+1)); done
+         [ "$_cshown" != "$_nsk $NCMD_SHOWN" ]; then fail "start.sh would print '$_cshown' skills/commands; the README gates count $_nsk and $NCMD_SHOWN"
+    else pass "start.sh prints the README's counts ($_cshown skills/commands; fixture 1 1 with the experimental one left out)"; fi
+  fi
   # Crewforth installs through npx, the Claude Code plugin or the release archive; the Homebrew channel was
   # removed. A line that tells a reader to `brew install` Crewforth, or names the old tap, points at a formula
   # nobody maintains. `brew install node` and other tools stay: those install something else. The CHANGELOG is
@@ -2081,7 +2100,7 @@ sec "== 6f) always-on token budget =="
 # for that cost, and a gate rather than a reminder — a verbose new description fails the suite instead of
 # quietly taxing every future session. Budgets sit just above the current sizes: raising one is allowed, but
 # only as a deliberate edit here.
-BUDGET_DISC=13711    # 5R.3: 13712 → 13711 (the old-name phrases rewritten, net −1). 5d.2 prompt audit: tightened to the measured size (13719 → 13712: format-to-content style line, one reload
+BUDGET_DISC=13741    # RC-1: 13711 → 13741 (+30): "Reply in the user's language even when a skill's text is English" — a Turkish session answered /crew-review in English, because every skill body it had just read was English; the rule is stated nowhere else. evals/cases/reply-language measures it. 5R.3: 13712 → 13711 (the old-name phrases rewritten, net −1). 5d.2 prompt audit: tightened to the measured size (13719 → 13712: format-to-content style line, one reload
                      # answer, the orphaned background-warning line removed). Before that: 3.0 rename (suffix → crew- prefix): +23 B (23 occurrences), not content — measured 13696 → 13719.
                      # DISCIPLINE.md (the discipline half of CLAUDE.md); before 3.0 the ceiling was 13700, currently 13601. (2026-09-18, a second
                      # +100 B on top of the raise below, and the whole of it went into ONE sentence of §4.6: a commit
@@ -2178,53 +2197,17 @@ BUDGET_AGENTS=5596   # 5d.2: tightened to the measured sum (was 5800 with 204 B
                      # +crew-performance-expert (~426B) — security, privacy and tests each had an independent
                      # reviewer and performance was the one quality axis where the author audited their own
                      # work. Bought at ~110 tokens per session; the alternative was leaving that gap open.)
-BUDGET_SKILLS=10249 # 5S: 10255 → 10249 (crew-doctor: "the install"). 5R.4: 10259 → 10255 (crew-update: "a newer version"). 5d.2: tightened to the measured size (10265 → 10259: two descriptions lost a stale word). Before that:
-                    # 3.0: commands merged into skills — 661 B previously in the listing but uncounted, not new content:
-                    # the six model-invocable commands (doctor 136 · handoff 91 · plan 151 · review 82 · ship 78 ·
-                    # update 123) now live in skills/, where this sum sees them. The five user-only ones
-                    # (disable-model-invocation: true) are not in the listing and are not counted. The other skills
-                    # measured 9604 before and after — the ceiling moved by exactly the 661, not a byte for anything new.
-                    # Before that: 9604 — 3.0 rename (suffix → crew- prefix): +7 B (7 occurrences), not content — 9597 → 9604.
-                    # Before 3.0: 9600; sum of skill frontmatter; currently 9597 — **3 bytes of headroom**, measured 2026-09-23 from
-                    # this suite's own line. 3.0 swapped cqrs-aop-module (-202 B) for backend-architecture (+211 B) and
-                    # the ceiling was NOT raised: the new description was cut until it fit. Before that 9588, measured
-                    # 2026-09-20 (the note said 9521 and was 67 B stale). Read that margin before editing any
-                    # description: one added clause trips this gate, and that is the ratchet working, not a bug.
-                    # Raising the ceiling needs the same thing every bump here needed — a written reason for what
-                    # the bytes buy. (2.6.x: +843 B — ten descriptions gained a
-                    # "Use when …" sentence. The field's job is to say WHEN to reach for the skill; a description
-                    # that only says what its author knows is matched by nothing, and inside Crewforth that was
-                    # invisible because route-hint.sh and the trigger map do the routing. Outside the harness —
-                    # a skill copied into another project, another client, a bare session — the routing is gone
-                    # and the description is all there is. STATED HONESTLY: this bump does NOT fix the small-window
-                    # case. The listing budget is 1% of the context window, so a 200k model allows ~2,000 B and
-                    # Crewforth is far past that with or without these ten sentences. What changed is that the
-                    # remedy is now targetable: `eval/utilization.sh` reports which skills nothing in a project
-                    # actually reached, which is the list `skillOverrides: name-only` needs and never had. The
-                    # next skill that wants room takes it from a description, not from another bump.)
-                    # Was 8700 / 8678. (2.5.0: +307 B for `automode-policy` — about
-                    # half of it the allowed-tools grant that lets the skill run its own verifier without a
-                    # prompt, not prose. The listing is what every session pays for; the next skill that wants
-                    # room takes it from a description, not from another bump.) Was 8380 / 8371 — nine bytes
-                    # of headroom, so the ceiling was already the binding constraint, not this skill. (2.3.0: +187 B for `teamboard`. It is the only
-                     # skill whose absence from the LISTING is silently unsafe rather than merely unhelpful: a
-                     # skill that fails to match usually means the model does the work itself, but this one
-                     # failing to match means two people do the SAME work, on separate machines, discovering it
-                     # at merge time. Trimmed to one sentence and the words people actually type — claim, item,
-                     # team board.) RATCHETED DOWN in 1.11.0 from 12,350: the
-                     # `Trigger phrases:` lines moved out of every skill's `description` into the body. This is not
-                     # cosmetic. Claude Code loads a LISTING of skill names+descriptions every session and the
-                     # budget is 1% of the context window; over it, descriptions are truncated or dropped outright,
-                     # "which can strip the keywords Claude needs to match your request" (official skills docs). The
-                     # Crewforth's listing was 11,372 chars against a 10,000 budget on a 1M window — overflowing on every
-                     # model, and 5.7x over on a 200k one. Now 7,208. A kit whose own skills push its descriptions
-                     # out of the listing is a kit that stops matching, which is exactly the symptom users report.
-                     # (1.8.0: +confidence-check (~359B), Crewforth's
-                     # only gate that fires BEFORE implementation — every other one reviews code that already
-                     # exists, and none catch correct code that should never have been written; and
-                     # +dependency-upgrade (~444B), split from dependency-audit because one reports and the
-                     # other rewrites lockfiles: different risk, different DoD, and an audit you can run on any
-                     # branch stops being safe the moment it can also apply things)
+BUDGET_SKILL_LISTING=9063 # the skill listing as Claude Code counts it (eval/lib/skill-listing.awk — the count doctor.sh
+                          # reports too), in CHARACTERS: 46 model-invocable skills, 9063, the figure Claude Code's own
+                          # "Skill listing over budget" warning printed for a fresh install. It replaced a byte sum of
+                          # the frontmatter that no tool used (10249), so doctor and this gate now read one number.
+                          # The ceiling is "does not grow", not a target. 8,000 was weighed and rejected: that is the
+                          # whole listing's budget on a 200k model at the default fraction, and Claude Code's own
+                          # skills take ~5,900 of it (v2.1.282), so no Crewforth-only figure could make the whole list
+                          # fit. The full install raises the fraction to 0.04; the plugin cannot, so doctor and
+                          # /crew-doctor tell a plugin user on a 200k model the one line to add and what it costs.
+                          # Shorter descriptions wait for an eval that measures whether Claude still picks the skill
+                          # from them — trimming blind trades a number for matching nobody measured.
 # A SKILL.md's `metadata:` block is Crewforth's own catalogue data (`kind: command`); Claude Code does not act on it
 # and it never enters the listing, so it is not counted. Nothing else in a frontmatter is skipped.
 fm_bytes(){ awk '/^---$/{c++; next} c==1 { if ($0 ~ /^metadata:/) { m=1; next } if (m && $0 ~ /^[ \t]/) next; m=0; print }' "$1" 2>/dev/null | wc -c | tr -d ' '; }
@@ -2242,15 +2225,33 @@ elif [ -f "$ROOT/DISCIPLINE.md" ]; then
   DB="$(wc -c < "$ROOT/DISCIPLINE.md" | tr -d ' ')"; DBCR="$(tr -dc '\r' < "$ROOT/DISCIPLINE.md" | wc -c | tr -d ' ')"
 else DB=0; DBCR=0; fi
 AB=0; for f in "$AGENTS"/*.md;      do [ -e "$f" ] && AB=$((AB + $(fm_bytes "$f"))); done
-# THE RULE: a skill counts unless its frontmatter says `disable-model-invocation: true`. Per the skills reference,
-# with that flag "the description is not in context" — only the user can invoke it, so it costs the listing nothing.
-# Every other skill's description IS in the listing every session, and that is what this budget is for. (Before
-# 3.0 the six model-invocable commands sat in the same listing from commands/, and this sum never saw them.)
-SB=0; for f in "$SKILLS"/*/SKILL.md; do
-  [ -e "$f" ] || continue
-  grep -q '^disable-model-invocation:[[:space:]]*true' "$f" && continue
-  SB=$((SB + $(fm_bytes "$f")))
-done
+# The skill listing is counted by eval/lib/skill-listing.awk, the file doctor.sh uses: a skill with
+# `disable-model-invocation: true` is not listed, every other one costs its name and description as Claude Code counts
+# them. The counter is pinned first on the fixtures Claude Code itself was measured on, so a drift in the awk (or in an
+# awk: BSD, mawk and Git Bash gawk all run this) reads as a broken meter, not as a budget.
+SLA="$ROOT/eval/lib/skill-listing.awk"; [ -f "$SLA" ] || SLA="$SGR/kit/eval/lib/skill-listing.awk"
+if [ ! -f "$SLA" ]; then fail "eval/lib/skill-listing.awk is missing — the skill listing cannot be counted"; SL=0
+else
+  _slf="$(mktemp -d)"; mkdir -p "$_slf/a" "$_slf/b" "$_slf/c" "$_slf/d"
+  printf -- '---\nname: aa\ndescription: bbb\n---\n' > "$_slf/a/SKILL.md"
+  printf -- '---\nname: cc\ndescription: dddd\n---\n' > "$_slf/b/SKILL.md"
+  printf -- '---\nname: ee\ndescription: |\n  one\n  two\n---\n' > "$_slf/c/SKILL.md"
+  printf -- '---\nname: ff\ndescription: gg\ndisable-model-invocation: true\n---\n' > "$_slf/d/SKILL.md"
+  _c1="$(LC_ALL=C awk -f "$SLA" "$_slf/a/SKILL.md")"; _c2="$(LC_ALL=C awk -f "$SLA" "$_slf/a/SKILL.md" "$_slf/b/SKILL.md")"
+  _c3="$(LC_ALL=C awk -f "$SLA" "$_slf/c/SKILL.md")"; _c4="$(LC_ALL=C awk -f "$SLA" "$_slf/a/SKILL.md" "$_slf/d/SKILL.md")"
+  _c5="$(printf -- '---\nname: hh\ndescription: a\342\200\224b\n---\n' > "$_slf/e.md"; LC_ALL=C awk -f "$SLA" "$_slf/e.md")"
+  rm -rf "$_slf"
+  # 9 and 20 are Claude Code's own readings; ee = 2 + len("onetwo") + 4 (a line break costs nothing, as measured on
+  # 46 shipped skills); ff is not listed; an em dash is one character (hh = 2 + 3 + 4).
+  if [ "$_c1" = "9 1 0" ] && [ "$_c2" = "20 2 0" ] && [ "$_c3" = "12 1 0" ] && [ "$_c4" = "9 1 0" ] && [ "$_c5" = "9 1 0" ]; then
+    pass "skill-listing.awk reproduces Claude Code's counter on its fixtures (9 · 20 · block lines · unlisted · em dash)"
+  else fail "skill-listing.awk drifted from Claude Code's counter: got '$_c1' '$_c2' '$_c3' '$_c4' '$_c5', want '9 1 0' '20 2 0' '12 1 0' '9 1 0' '9 1 0'"; fi
+  read -r SL SLN SLU <<EOF_SL
+$(LC_ALL=C awk -f "$SLA" "$SKILLS"/*/SKILL.md)
+EOF_SL
+  [ "${SLU:-0}" = 0 ] && pass "no skill uses a listing shape the count was not measured on (when_to_use, folded description)" \
+    || fail "$SLU skill(s) use when_to_use or a folded description — measure that shape against Claude Code before trusting the listing count"
+fi
 # The budget GATES Crewforth's payload (kit repo, IS_KIT). In an INSTALLED project the user's own agents/skills —
 # including the ones adopt imports from a taken-over agent — legitimately add to the always-on cost (their choice),
 # so there we REPORT the numbers instead of failing the suite.
@@ -2263,20 +2264,21 @@ done
 #
 # $4 is the number of carriage returns in the same text the budget was measured on, so the two numbers describe
 # the same bytes. A real overrun still says "over budget"; only a CRLF one is renamed.
-bud(){ # $1 name  $2 measured  $3 budget  $4 (optional) carriage returns in the measured text
-       if [ "$2" -le "$3" ]; then pass "$1 within budget ($2 ≤ $3 bytes)"
+bud(){ # $1 name  $2 measured  $3 budget  $4 (optional) carriage returns in the measured text  $5 (optional) unit
+       local u="${5:-bytes}"
+       if [ "$2" -le "$3" ]; then pass "$1 within budget ($2 ≤ $3 $u)"
        elif [ "$IS_KIT" = 1 ]; then
          local cr="${4:-0}"
          if [ "$cr" -gt 0 ] && [ $(( $2 - cr )) -le "$3" ]; then
            fail "$1 over budget ONLY because this checkout is CRLF: $2 > $3 bytes, and $cr of those bytes are carriage returns ($(( $2 - cr )) with LF endings, which is within budget). Re-check out the file rather than editing the budget."
          else
-           fail "$1 over budget: $2 > $3 bytes"
+           fail "$1 over budget: $2 > $3 $u"
          fi
-       else pass "$1 $2 bytes (over Crewforth's $3 baseline — your project's own additions, not gated in an install)"; fi; }
+       else pass "$1 $2 $u (over Crewforth's $3 baseline — your project's own additions, not gated in an install)"; fi; }
 bud "discipline"         "$DB" "$BUDGET_DISC" "$DBCR"
 bud "agent descriptions" "$AB" "$BUDGET_AGENTS"
-bud "skill descriptions" "$SB" "$BUDGET_SKILLS"
-echo "   always-on total: $((DB+AB+SB)) bytes (budget $((BUDGET_DISC+BUDGET_AGENTS+BUDGET_SKILLS)))"
+bud "skill listing" "$SL" "$BUDGET_SKILL_LISTING" 0 chars
+echo "   always-on: discipline + agents $((DB+AB)) bytes (budget $((BUDGET_DISC+BUDGET_AGENTS))) · skill listing $SL chars for ${SLN:-0} skills (budget $BUDGET_SKILL_LISTING)"
 # The diagnosis is pinned, not only written, on CRLF copies of this very file: the carriage-return count must cover
 # exactly the measured lines, a copy whose text is exactly at the budget with LF endings must be named as CRLF with its
 # figures, and one a byte past the budget must still read "over budget". All three failed while the count came from
@@ -4026,6 +4028,30 @@ case "$o" in *'"additionalContext"'*SESSION_STATE*) pass "handover present -> in
   *) fail "session-rehydrate did not inject a pointer when SESSION_STATE.md exists" ;; esac
 if [ -n "$JSONQ" ]; then printf '%s' "$o" | json_ok && pass "rehydrate output is valid JSON ($JSONQ)" || fail "rehydrate output is not valid JSON";
     else skip tool "the rehydrate output JSON-validity check (no working jq)"; fi
+# The install language (RC-2 rehearsal: a bare /crew-doctor in a Turkish install came back in English). The hook
+# now runs on startup too and tells the model the install language; the handover is still offered only on a
+# boundary. Six states, each read by what the hook actually printed: a Turkish install gets the language line on
+# startup and on a boundary; an English install gets none; startup never offers the handover, a boundary does.
+# JSON validity is asked of Crewforth's own reader, so it holds on a machine without jq.
+_rh(){ printf '{"hook_event_name":"SessionStart","cwd":"%s","source":"%s"}' "$RHD" "$1" | CLAUDE_PROJECT_DIR= bash "$HOOKS/session-rehydrate.sh" 2>/dev/null; }
+_rk(){ case "$1" in *'handover'*) printf 'H' ;; esac; case "$1" in *'installed in Turkish'*) printf 'L' ;; esac; }
+mkdir -p "$RHD/.claude"
+printf 'lang=tr\n' > "$RHD/.claude/kit.conf"; _o1="$(_rh startup)"; _o2="$(_rh compact)"
+printf 'lang=en\n' > "$RHD/.claude/kit.conf"; _o3="$(_rh startup)"; _o4="$(_rh compact)"
+rm -f "$RHD/docs/SESSION_STATE.md"; _o5="$(_rh startup)"
+printf 'lang=tr\n' > "$RHD/.claude/kit.conf"; _o6="$(_rh clear)"
+_got="$(_rk "$_o1")/$(_rk "$_o2")/$(_rk "$_o3")/$(_rk "$_o4")/$(_rk "$_o5")/$(_rk "$_o6")"
+[ "$_got" = "L/HL//H//L" ] && pass "SessionStart names a Turkish install's language on startup and on a boundary; English adds nothing; startup never offers the handover (6 states)" \
+  || fail "session-rehydrate language/handover states read '$_got', want 'L/HL//H//L' (tr startup · tr compact+handover · en startup · en compact+handover · en startup no handover · tr clear no handover)"
+_rj=0; for _o in "$_o1" "$_o2" "$_o4" "$_o6"; do printf '%s' "$_o" > "$RHD/o.json"; awk -v op=validate -f "$ROOT/eval/lib/settings-json.awk" "$RHD/o.json" 2>/dev/null || _rj=$((_rj+1)); done
+[ "$_rj" = 0 ] && pass "each language/handover output is valid JSON (Crewforth's reader, 4 of 4)" || fail "$_rj session-rehydrate output(s) are not valid JSON"
+# Asked of the entry that carries session-rehydrate, not of the file: skill-trust's entry has the same matcher.
+_rw(){ _n="$(awk -v op=len -v path=hooks.SessionStart -f "$ROOT/eval/lib/settings-json.awk" "$1" 2>/dev/null)"; _i=0
+       while [ "$_i" -lt "${_n:-0}" ]; do _e="$(awk -v op=get -v path="hooks.SessionStart.$_i" -f "$ROOT/eval/lib/settings-json.awk" "$1" 2>/dev/null)"
+         case "$_e" in *session-rehydrate*) case "$_e" in *'"matcher"'*startup*) return 0 ;; esac ;; esac; _i=$((_i+1)); done; return 1; }
+sed 's/"startup|resume|clear|compact|fork"/"compact|clear|resume|fork"/' "$ROOT/settings.json" > "$RHD/s-twin.json"
+if _rw "$ROOT/settings.json" && ! _rw "$RHD/s-twin.json"; then pass "settings.json runs session-rehydrate on startup too; a copy wired as before (no startup) is caught"
+else fail "session-rehydrate is not wired on startup (or the check cannot tell) — a Turkish install's first session would not be told its language"; fi
 rm -rf "$RHD"
 # board-sync builds its JSON with one awk escaper on every machine. The no-jq escaper it replaced handled only the
 # quote, the backslash and the newline, so a TAB in an item title produced JSON the CLI cannot parse and the board
@@ -4148,7 +4174,7 @@ mkdir -p "$DCR/.claude/skills/only-mine" "$DCR/.claude/hooks"
 cp -R "$SKILLS/handoff" "$DCR/.claude/skills/" 2>/dev/null
 printf '# x\n' > "$DCR/.claude/skills/only-mine/SKILL.md"
 cp "$HOOKS"/*.sh "$DCR/.claude/hooks/" 2>/dev/null; chmod +x "$DCR/.claude/hooks/"*.sh 2>/dev/null
-_own(){ ( cd "$DCR" && bash "$ROOT/eval/doctor.sh" 2>&1 | grep -oE '[0-9]+ project-specific skill' | head -1 | cut -d' ' -f1 ); }
+_own(){ ( cd "$DCR" && CREW_LANG=en bash "$ROOT/eval/doctor.sh" 2>&1 | grep -oE '[0-9]+ project-specific skill' | head -1 | cut -d' ' -f1 ); }
 printf 'skills/handoff\n'   > "$DCR/.claude/kit-manifest.txt"; _lf="$(_own)"
 printf 'skills/handoff\r\n' > "$DCR/.claude/kit-manifest.txt"; _crlf="$(_own)"
 [ -n "$_lf" ] && [ "$_lf" = "$_crlf" ] \
@@ -4191,7 +4217,7 @@ bash "$ROOT/eval/doctor.sh" "$DOC" >/dev/null 2>&1 \
   && pass "doctor: CLAUDE.md with the @import -> exit 0" \
   || fail "doctor flagged a CLAUDE.md that DOES import the discipline"
 # Readiness is ADVISORY: a bare project trips every readiness signal, and the verdict must stay exit 0.
-DOUT="$(bash "$ROOT/eval/doctor.sh" "$DOC" 2>&1)"; DRC=$?
+DOUT="$(CREW_LANG=en bash "$ROOT/eval/doctor.sh" "$DOC" 2>&1)"; DRC=$?
 [ "$DRC" -eq 0 ] && pass "doctor: readiness gaps do NOT change the verdict (advisory)" || fail "readiness gaps changed doctor's exit code — it must stay a statement about the install"
 case "$DOUT" in *"Readiness (advisory"*) pass "doctor prints the readiness block" ;; *) fail "doctor readiness block missing" ;; esac
 case "$DOUT" in *"➖"*) pass "readiness flags gaps on a bare project (devcontainer/MCP/manifest absent)" ;; *) fail "readiness found no gap on a bare project — the signals are not firing" ;; esac
@@ -5468,13 +5494,13 @@ if [ -f "$GR" ]; then
   DTMP="$(mktemp -d)"; mkdir -p "$DTMP/.claude"
   for d in eval hooks skills agents; do [ -d "$ROOT/$d" ] && cp -R "$ROOT/$d" "$DTMP/.claude/$d"; done
   cp "$ROOT/settings.json" "$DTMP/.claude/settings.json" 2>/dev/null
-  DOUT="$(cd "$DTMP" && bash .claude/eval/doctor.sh 2>"$DTMP/err")"
+  DOUT="$(cd "$DTMP" && CREW_LANG=en bash .claude/eval/doctor.sh 2>"$DTMP/err")"
   # An install from before 2.5.0 keeps the old `Bash`-only matcher, and nothing in the session looks wrong
   # while every PowerShell command walks past §4.5. doctor has to SAY so, so this drives the downgrade.
   case "$DOUT" in *"watch both Bash and PowerShell"*) pass "doctor confirms the shell matcher covers PowerShell" ;;
                   *) fail "doctor did not report on the shell matcher" ;; esac
   sed 's/"Bash|PowerShell"/"Bash"/' "$DTMP/.claude/settings.json" > "$DTMP/s.tmp" && mv "$DTMP/s.tmp" "$DTMP/.claude/settings.json"
-  DOUT2="$(cd "$DTMP" && bash .claude/eval/doctor.sh 2>/dev/null)"
+  DOUT2="$(cd "$DTMP" && CREW_LANG=en bash .claude/eval/doctor.sh 2>/dev/null)"
   case "$DOUT2" in *"watch only Bash"*) pass "doctor flags a pre-2.5.0 Bash-only matcher as a failure" ;;
                    *) fail "doctor stayed quiet on a Bash-only matcher — the gap is invisible to an upgrader" ;; esac
   grep -q 'command not found' "$DTMP/err" && fail "doctor.sh calls a helper before it is defined (see stderr)" \
@@ -5487,9 +5513,38 @@ if [ -f "$GR" ]; then
   # that has it, so it only ever exercised one of doctor's four branches; CI has no CLI, took the fourth, and
   # failed on a wording difference. Running it both ways is what makes the assertion about doctor rather than
   # about the machine the suite happens to run on.
-  DOUT3="$(cd "$DTMP" && PATH=/usr/bin:/bin bash .claude/eval/doctor.sh 2>/dev/null)"
+  DOUT3="$(cd "$DTMP" && CREW_LANG=en PATH=/usr/bin:/bin bash .claude/eval/doctor.sh 2>/dev/null)"
   case "$DOUT3" in *"auto-mode classifier"*) pass "doctor reports the auto-mode line with no claude CLI present" ;;
                    *) fail "doctor went silent on auto-mode when the claude CLI is absent" ;; esac
+  # (f) doctor speaks the install's language. A Turkish install ran a Turkish installer and then got an English
+  #     doctor (RC-1 rehearsal); the language now comes from the `lang=` start.sh/adopt.sh record in kit.conf. Four
+  #     claims, driven on this fixture: the record picks the language with no variable set; no line the doctor
+  #     printed fell back to English (CREW_I18N_MISS names each key that had no row); every identifier the English
+  #     run printed — a path, a command, a settings key — is still there, untranslated, in the Turkish run; and an
+  #     English record gets English. The must-fail twin plants a line with no row in a copy of the doctor and
+  #     needs the miss list to name it, so an empty list means "all translated", not "the collector is dead".
+  _dl(){ printf 'lang=%s\n' "$1" > "$DTMP/.claude/kit.conf"; }
+  _dl tr; : > "$DTMP/miss"
+  DTR="$(cd "$DTMP" && env -u CREW_LANG CREW_I18N_MISS="$DTMP/miss" bash .claude/eval/doctor.sh 2>/dev/null)"
+  _dl en; DEN="$(cd "$DTMP" && env -u CREW_LANG bash .claude/eval/doctor.sh 2>/dev/null)"
+  case "$DTR" in *"kurulum denetimi"*) pass "doctor speaks Turkish when the install recorded lang=tr (no variable set)" ;;
+                 *) fail "doctor ignored lang=tr in kit.conf — a Turkish install gets an English doctor" ;; esac
+  case "$DEN" in *"install doctor"*) pass "doctor speaks English when the install recorded lang=en" ;;
+                 *) fail "doctor did not speak English for lang=en" ;; esac
+  [ ! -s "$DTMP/miss" ] && pass "every line the Turkish doctor printed has a translation (miss list empty)" \
+    || fail "the Turkish doctor fell back to English for: $(tr '\n' '|' < "$DTMP/miss")"
+  _idm=""; _idn=0
+  for _id in $(printf '%s\n' "$DEN" | grep -oE '\.claude/[A-Za-z0-9_./*-]+|settings\.json|core\.hooksPath|/crew-[a-z-]+|npx crewforth [a-z]+|skillListingBudgetFraction|PreToolUse|UserPromptSubmit|SessionStart' | sort -u); do
+    _idn=$((_idn + 1)); case "$DTR" in *"$_id"*) ;; *) _idm="$_idm $_id" ;; esac
+  done
+  if [ "$_idn" -lt 5 ]; then fail "identifier check read only $_idn identifier(s) from the English doctor — the extractor is broken, not the doctor"
+  elif [ -z "$_idm" ]; then pass "the Turkish doctor keeps all $_idn identifiers of the English one untranslated (paths, commands, keys)"
+  else fail "the Turkish doctor lost or translated identifier(s):$_idm"; fi
+  sed 's/^_mt "== Crewforth — install doctor =="; echo "$_M"$/&; ok "planted line with no translation"/' "$DTMP/.claude/eval/doctor.sh" > "$DTMP/.claude/eval/doctor-twin.sh"
+  _dl tr; : > "$DTMP/miss2"
+  ( cd "$DTMP" && env -u CREW_LANG CREW_I18N_MISS="$DTMP/miss2" bash .claude/eval/doctor-twin.sh >/dev/null 2>&1 )
+  grep -qx 'planted line with no translation' "$DTMP/miss2" && pass "twin: a planted untranslated line is named by the miss list" \
+    || fail "twin: the miss list did not name a planted untranslated line — the empty list above proves nothing"
   rm -rf "$DTMP" "$GTMP"
 else
   fail "eval/gate-report.sh missing from the payload"
@@ -6237,10 +6292,12 @@ if [ "$IS_KIT" = 1 ]; then
     command -v node >/dev/null 2>&1 && node "$KR/bin/cli.js" --help; } > "$_kd/out.txt" 2>&1
   _kc="$(kw_lines "$_kd/out.txt" | sed "s|^$_kd/||")"; _kn="$(grep -c . "$_kd/out.txt")"
   # Twins. Must fail: a start.sh whose table says "full kit" again, and output that says "installing Crewforth.".
-  # Must pass: output that only names paths — .claude/kit.conf, kit/ deleted, kit-manifest.txt, drizzle-kit, KIT_X.
+  # Must pass: output that only names paths — .claude/kit.conf, the folder in the archive's own error, kit-manifest.txt,
+  # drizzle-kit, KIT_X. ("kit/ deleted" used to sit here as a path; it was the last line of every install, and the
+  # RC-1 rehearsal read it as the old word. The closing lines are held to it separately below.)
   sed 's/"full install") s=/"full kit") s=/' "$KR/start.sh" > "$_kd/mut-start.sh"
   _ok=kit; printf 'Done: installing the %s.\nKurulum: %sin dosyaları hazır\n' "$_ok" "$_ok" > "$_kd/bad.txt"
-  printf 'wrote .claude/kit.conf\nkit/ deleted\nsee .claude/kit-manifest.txt and drizzle-kit status\nKIT_X=1\n' > "$_kd/good.txt"
+  printf 'wrote .claude/kit.conf\nERROR: kit/ not found\nsee .claude/kit-manifest.txt and drizzle-kit status\nKIT_X=1\n' > "$_kd/good.txt"
   if [ -n "$_ka$_kb$_kc" ]; then fail "a printed line still says kit — say Crewforth:
 $(printf '%s\n%s\n%s\n' "$_ka" "$_kb" "$_kc" | grep . | head -n 6 | sed "s|$KR/||; s|^|       |")"
   elif [ "$_kn" -lt 30 ]; then fail "FIXTURE: the captured help and preflight output is only $_kn line(s) — the run broke, not the wording"
@@ -6248,6 +6305,16 @@ $(printf '%s\n%s\n%s\n' "$_ka" "$_kb" "$_kc" | grep . | head -n 6 | sed "s|$KR/|
   elif [ "$(kw_lines "$_kd/bad.txt" | grep -c .)" != 2 ]; then fail "Crewforth-word matcher missed a planted old-name word (English or Turkish) in the output"
   elif [ -n "$(kw_lines "$_kd/good.txt")" ]; then fail "Crewforth-word matcher flagged a path or identifier: $(kw_lines "$_kd/good.txt" | head -n 2 | tr '\n' ' ')"
   else pass "no printed line says kit: terminal-only scripts, the installers' tables/echo/usage and cli --help, plus $_kn lines of real help and preflight output (EN+TR); a reverted table entry and planted words are caught, paths are not"; fi
+  # The closing line of an install is the one line every user reads. It may name no kit at all, not even as a path:
+  # the key and its Turkish value are both read, and a planted "kit/ deleted" in a copy must be caught.
+  _dn(){ grep -hE '"Done\.|'"'"'Done\.' "$@" 2>/dev/null | grep -iE '(^|[^a-z_.-])kit([^a-z_-]|$)'; }
+  if [ ! -f "$SGR/start.sh" ]; then :
+  elif [ -n "$(_dn "$SGR/start.sh" "$SGR/adopt.sh")" ]; then fail "an installer's closing line still names kit: $(_dn "$SGR/start.sh" "$SGR/adopt.sh" | head -n 1 | cut -c1-120)"
+  else
+    sed 's/the installer files are removed\./kit\/ deleted./' "$SGR/start.sh" > "$_kd/done-twin.sh"
+    [ -n "$(_dn "$_kd/done-twin.sh")" ] && pass "the installers' closing lines name no kit (EN key and TR value); a planted 'kit/ deleted' is caught" \
+      || fail "twin: the closing-line check missed a planted 'kit/ deleted' — it reads nothing"
+  fi
   rm -rf "$_kd"
 else
   skip scope "printed-text wording check skipped (installed project — the installers live in the source repository)"

package/kit/hooks/guard-bash.sh

@@ -1390,7 +1390,9 @@ if git_has "$CMD" 'commit|push'; then
       # §4.4 branch guard: committing straight onto main/master is not blocked (a fresh project legitimately
       # lives on main), but it is surfaced in the approval prompt so the user can send it to a branch instead.
       BRANCH_WARN=""
-      case "$(git rev-parse --abbrev-ref HEAD 2>/dev/null)" in
+      # symbolic-ref names the branch even before its first commit, where rev-parse answers "HEAD" and the very
+      # first commit onto main went by without this line. A detached HEAD reads empty: not the default branch.
+      case "$(git symbolic-ref --short -q HEAD 2>/dev/null)" in
         main|master) BRANCH_WARN="⚠️  This commits DIRECTLY to the default branch. Prefer a feature branch unless you meant to.
 
 " ;;

package/kit/hooks/session-rehydrate.sh

@@ -8,8 +8,14 @@
 #   - SessionStart is the right event: its stdout / hookSpecificOutput.additionalContext is injected into the
 #     model's context (docs: "Re-inject context after compaction"). PreCompact is NOT used — it can only block
 #     compaction, not add instructions to the summary, and blocking is not what we want.
-#   - Matched on compact|clear|resume, NOT startup: those three are explicit "continue where I left off" boundaries.
-#     startup fires on every unrelated session open and would keep surfacing a stale handover — that would nag.
+#   - The HANDOVER is offered on compact|clear|resume|fork only, NOT on startup: those are explicit "continue where I
+#     left off" boundaries, while startup fires on every unrelated session open and would keep surfacing a stale
+#     handover — that would nag. The hook itself also runs on startup, for the language line below; it reads the
+#     payload's `source` to tell the two apart (no `source` -> treated as a boundary, the pre-3.0 payload shape).
+#   - THE LANGUAGE LINE. A project installed in Turkish (`lang=tr` in .claude/kit.conf) is told so at every session
+#     start: reply in Turkish unless the user writes in another language. The RC-2 rehearsal typed a bare
+#     `/crew-doctor` in a Turkish install: the doctor printed Turkish and the summary came back in English, because a
+#     bare command has no word to take the language from. An English install adds nothing (English is the default).
 #   - CLAUDE.md itself reloads on /compact and /clear on its own, so this hook does NOT re-inject the discipline —
 #     only the session-specific state the reload can't recover.
 #   - Fails OPEN and SILENT: no handover file, no output. It never blocks the session (always exits 0).
@@ -28,13 +34,26 @@ fi
 # Undo the JSON escaping first (`\\`), then fold any lone separator. Both are no-ops on a POSIX path.
 ROOT="${ROOT//\\\\//}"; ROOT="${ROOT//\\//}"
 
-STATE="$ROOT/docs/SESSION_STATE.md"
-[ -s "$STATE" ] || exit 0   # no (non-empty) handover → nothing to rehydrate, stay silent
+SRC=""
+case "$IN" in *'"source"'*) SRC="${IN#*\"source\"}"; SRC="${SRC#*\"}"; SRC="${SRC%%\"*}" ;; esac
+
+LANGMSG=""
+if [ -f "$ROOT/.claude/kit.conf" ]; then
+  while IFS= read -r _l || [ -n "$_l" ]; do
+    case "${_l%$'\r'}" in lang=tr) LANGMSG="This project was installed in Turkish (lang=tr in .claude/kit.conf). Reply in Turkish unless the user writes in another language, including after a bare command such as /crew-doctor. Commands, paths, file names and identifiers stay as written." ;; esac
+  done < "$ROOT/.claude/kit.conf"
+fi
 
+MSG=""
+STATE="$ROOT/docs/SESSION_STATE.md"
+if [ "$SRC" != startup ] && [ -s "$STATE" ]; then
 MSG="A session handover from before this context boundary exists at docs/SESSION_STATE.md. If the user's request continues that work, read it first — it holds the in-progress task state, open decisions, and the intended next step — and do not restart from scratch. If the request is about something else, do not bring it up."
+fi
+[ -n "$LANGMSG" ] && MSG="${MSG:+$MSG }$LANGMSG"
+[ -n "$MSG" ] || exit 0   # no handover to offer and nothing to say about language -> stay silent
 
 # hookSpecificOutput.additionalContext is the documented channel that injects text into the model's context.
-# ONE PATH, no jq: MSG is a fixed constant with no quote, backslash, control character or newline, so it needs no
+# ONE PATH, no jq: MSG is built from fixed constants with no quote, backslash, control character or newline, so it needs no
 # escaping and every machine emits the same bytes. If you ever make MSG dynamic, escape it the way board-sync.sh
 # does (jq-identical, measured) — a raw quote or tab here makes JSON the CLI silently drops.
 printf '{"hookSpecificOutput":{"hookEventName":"SessionStart","additionalContext":"%s"}}\n' "$MSG"

package/kit/hooks/skill-trust.sh

@@ -21,11 +21,15 @@
 # never fail to start because of this.
 set -uo pipefail
 HERE="$(cd "$(dirname "$0")" && pwd)"
-MODE=""
-case "${1:-}" in --trust) MODE=trust ;; esac
+MODE=""; ONE=""
+# --trust-one <skills/name | agents/name.md>: record ONE component as accepted, and nothing else. The updater uses it
+# for the skill it turned into a project skill itself (cqrs-aop-module, which Crewforth shipped until 3.0): flagging
+# the updater's own move as unvetted was the first thing an updated session said (RC-1 rehearsal). --trust would
+# have accepted every other foreign component too, including ones the user never looked at.
+case "${1:-}" in --trust) MODE=trust ;; --trust-one) MODE=one; ONE="${2:-}" ;; esac
 
 IN=""
-[ ! -t 0 ] && [ "$MODE" != trust ] && IN="$(cat 2>/dev/null || true)"
+[ ! -t 0 ] && [ -z "$MODE" ] && IN="$(cat 2>/dev/null || true)"
 ROOT="${CLAUDE_PROJECT_DIR:-}"
 [ -n "$ROOT" ] || ROOT="$(printf '%s' "$IN" | sed -n 's/.*"cwd"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' | head -1)"
 [ -n "$ROOT" ] || ROOT="$PWD"
@@ -58,6 +62,15 @@ digest(){
   else cksum "$1" 2>/dev/null | tr -s ' ' | cut -d' ' -f1,2 | tr ' ' '-'; fi
 }
 
+if [ "$MODE" = one ]; then
+  case "$ONE" in skills/*/*|agents/*/*) exit 1 ;; skills/?*) p="$CL/$ONE/SKILL.md" ;; agents/?*.md) p="$CL/$ONE" ;; *) exit 1 ;; esac
+  [ -f "$p" ] || exit 1
+  dg="$(digest "$p")"; [ -n "$dg" ] || exit 1
+  [ -f "$TRUST" ] || printf '# Components reviewed and accepted by the user. Regenerate with: bash skill-trust.sh --trust\n' > "$TRUST" 2>/dev/null || exit 1
+  grep -qxF "$dg $ONE" "$TRUST" 2>/dev/null || printf '%s %s\n' "$dg" "$ONE" >> "$TRUST"
+  exit 0
+fi
+
 # The component list: a skill is its SKILL.md, an agent is its file. Only those Crewforth does NOT ship —
 # without a manifest we cannot tell kit-owned from project-owned, so we stay silent rather than guess.
 [ -f "$MAN" ] || exit 0

package/kit/settings.json

@@ -93,7 +93,7 @@
     ],
     "SessionStart": [
       {
-        "matcher": "compact|clear|resume|fork",
+        "matcher": "startup|resume|clear|compact|fork",
         "hooks": [
           {
             "type": "command",

package/kit/skills/crew-doctor/SKILL.md

@@ -27,10 +27,20 @@ Verify Crewforth is actually *active* in this project (not just present on disk)
    it prints once per Crewforth version, and it is meant for the user, not for you.
 
 If `.claude/eval/doctor.sh` doesn't exist, this is not a full (start.sh / adopt.sh) install — Crewforth is likely
-running as a **plugin**, whose hooks are managed by Claude Code itself; there's nothing for the doctor to check.
+running as a **plugin**, whose hooks are managed by Claude Code itself; there's nothing for the doctor to check,
+with one exception: the skill listing, which a plugin cannot size for itself (its settings cannot set
+`skillListingBudgetFraction`). Count it the way the full install's doctor does:
+`LC_ALL=C awk -f "${CLAUDE_PLUGIN_ROOT}/eval/lib/skill-listing.awk" "${CLAUDE_PLUGIN_ROOT}"/skills/*/SKILL.md`
+(prints characters, then skills). At the default fraction a 200,000-token model has 8,000 characters for the whole
+listing, and Claude Code's own skills take about 5,900 of them (measured on v2.1.282). If Crewforth's number plus
+5,900 is over 8,000, tell the user plainly: on such a model the least-used skills lose their descriptions and stop
+being picked; one line in `~/.claude/settings.json` fixes it — `"skillListingBudgetFraction": 0.04`; the cost is
+that the listing is then sent whole every turn, about (that total ÷ 4) tokens, (tokens ÷ 2,000)% of a 200k window.
+It is their call — show the line and the cost, do not edit their settings. On a 1,000,000-token model the default
+already fits.
 
 **The eval scripts are installer-only, by decision.** `eval/` — `doctor.sh`, `smoke-test.sh`, `routing-eval.sh`,
 `scan-skill.sh`, `utilization.sh` — ships with `start.sh` and `adopt.sh` and NOT with the plugin edition. The eval
 scripts are developer instruments: they inspect an installation from outside it, and the plugin edition has no
-installation to inspect. Say this plainly when someone asks why `/crew-doctor` reports nothing on a plugin install,
+installation to inspect. Say this plainly when someone asks why `/crew-doctor` reports only the listing on a plugin install,
 rather than treating it as a defect. The Studio panel ships in both editions (`/crew-studio`).

package/kit/skills/crew-review/SKILL.md

@@ -14,10 +14,13 @@ Run the change set through the review trio (read-only). **The audits go out in p
    - @agent-crew-privacy-agent (`privacy-compliance`) — **when the diff touches personal data**: legal basis,
      minimisation, retention, transfer.
 2. (if SonarQube is in use) **sonarqube-check** — 0/0/0/0 gate (language-agnostic).
-3. **Last, and only once 1-2 are clean:** @agent-crew-review-agent (code-review) — "does it improve overall code
-   health"; severity-ranked comments. It is the closing reviewer in every writing agent's Coordination ("at
-   closure, report findings to crew-review-agent"), so it reads a diff the audits have already cleared — not the
-   other way round.
+3. **Last, once 1-2 leave no blocker:** @agent-crew-review-agent (crew-code-review) — "does it improve overall
+   code health": the plan (what changed, the risks), then findings with a severity and a category. "Clean" means
+   what crew-code-review means by it: **no critical or high finding open**. A medium or low audit finding does not
+   hold the reviewer back; hand it over in the reviewer's prompt and keep it in the report. A critical or high one
+   stops here: report it, leave the fix to its owner, and run the reviewer on the fixed diff. It is the closing
+   reviewer in every writing agent's Coordination ("at closure, report findings to crew-review-agent"), so it
+   reads a diff the audits have already cleared of blockers — not the other way round.
 
 Each agent returns a **short summary** to the main thread; raw output goes to `docs/` if needed. Do NOT modify code;
 collect findings in severity order, and leave the fix to the relevant expert.

package/kit/skills/crew-skill/SKILL.md

@@ -43,7 +43,8 @@ Skipping one of these is how a component ships half-installed:
    `node packaging/gen-diagrams.mjs` writes `assets/`, and smoke fails while any of them is stale.
 4. **Plugin edition** *(Crewforth repository only)*. `bash packaging/build-plugin.sh` — in the same commit, or the
    release stops at the sync gate.
-5. **Budget.** A new skill moves `BUDGET_SKILLS`; raise it in the same commit with the justification comment the
+5. **Budget.** A new skill moves `BUDGET_SKILL_LISTING` (the skill listing in characters, counted like Claude Code
+   counts it; `/crew-doctor` reports the same number). Raise it in the same commit with the justification comment the
    file's convention requires. Never raise it to make a red gate green without saying why.
 
 ## 4. Prove it — run all four, in this order

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "crewforth",
-  "version": "3.0.0-rc.1",
+  "version": "3.0.0-rc.3",
   "description": "Crewforth — your engineering crew for Claude Code: agents, skills and tool-level gates, in any stack.",
   "bin": {
     "crewforth": "bin/cli.js"

package/start.sh

@@ -123,16 +123,16 @@ _mt() {   # $1 = English text (the key); further args fill %s; result in _M
       "no effect:") s='etkisi yok:' ;;
       "Installing:") s='Kuruluyor:' ;;
       "Tip:  open Claude Code and run /crew-doctor — it checks the install is wired (hooks executable, core.hooksPath set, discipline imported) and scores the project's readiness. CLAUDE.md loads the discipline every session.") s="İpucu:  Claude Code'u açıp /crew-doctor çalıştırın — kurulumun eksiksiz bağlandığını denetler (hook'lar çalıştırılabilir mi, core.hooksPath ayarlı mı, disiplin import edilmiş mi) ve projenin ne kadar hazır olduğunu puanlar. Disiplin, CLAUDE.md sayesinde her oturumda yüklenir." ;;
-      "%s agents, %s skills installed.") s="%s ajan ve %s skill kuruldu." ;;
+      "%s agents, %s skills and %s commands installed.") s="%s ajan, %s skill ve %s komut kuruldu." ;;
       ".claude/DISCIPLINE.md written — owned by Crewforth; an update overwrites it, so keep your own rules out of it.") s=".claude/DISCIPLINE.md yazıldı. Bu dosya Crewforth'a ait ve her güncellemede yeniden yazılır; kendi kurallarınızı buraya eklemeyin." ;;
       "./CLAUDE.md created — EDIT the project section.") s='./CLAUDE.md oluşturuldu — proje bölümünü sizin DOLDURMANIZ gerekiyor.' ;;
       "trace scan: core.hooksPath -> .claude/hooks (§4.1/§4.2 commit gate active)") s='iz taraması: core.hooksPath -> .claude/hooks (§4.1/§4.2 commit kapısı açık)' ;;
-      "Done. ./.claude + ./CLAUDE.md ready (full install); kit/ deleted.") s='Tamamlandı. ./.claude ve ./CLAUDE.md hazır (tam kurulum); kit/ silindi.' ;;
+      "Done. ./.claude + ./CLAUDE.md ready (full install); the installer files are removed.") s='Tamamlandı. ./.claude ve ./CLAUDE.md hazır (tam kurulum); kurulum dosyaları silindi.' ;;
       "Next: 1) fill in the CLAUDE.md project section  2) open Claude Code at the repo root") s="Sıradaki adımlar: 1) CLAUDE.md'deki proje bölümünü doldurun  2) Claude Code'u deponun kökünde açın" ;;
       "Note: if Claude Code is ALREADY running here, restart it — CLAUDE.md and the discipline load at session start.") s='Not: Claude Code bu klasörde ZATEN açıksa yeniden başlatın — CLAUDE.md ve disiplin oturum açılırken yüklenir.' ;;
       "Panel: /crew-studio opens the Studio panel from this project (or: node .claude/studio/server/index.js --open).") s='Panel: /crew-studio komutu Studio panelini bu projeden açar (alternatif: node .claude/studio/server/index.js --open).' ;;
       "— backend + web + mobile (RN/Expo), every agent and skill") s="— backend, web ve mobil (RN/Expo); tüm ajanlar ve skill'ler" ;;
-      "%s agents · %s skills will be installed") s='%s ajan · %s skill' ;;
+      "%s agents · %s skills · %s commands will be installed") s='%s ajan · %s skill · %s komut' ;;
       "(shared: .claude/ and CLAUDE.md stay committable)") s="(paylaşımlı: .claude/ ve CLAUDE.md commit'lenebilir kalır)" ;;
       "(default — pass --shared to commit .claude/ and CLAUDE.md)") s="(varsayılan — .claude/ ve CLAUDE.md'yi commit'lemek için --shared verin)" ;;
       "no") s='hayır' ;;
@@ -548,13 +548,28 @@ count_installed() {   # $1=glob  -> count to install
   printf '%s' "$n"
 }
 N_AG="$(count_installed "$SRC/agents/*.md")"
-N_SK="$(count_installed "$SRC/skills/*/")"
+# Skills and commands the way the README and crewforth.com count them, so the install does not print a third number:
+# a skill whose metadata says `kind: command` is a command, one marked `experimental: true` is neither (it is
+# installed and works when typed, but nothing advertises it). Read from the frontmatter with builtins only.
+count_showcase() {   # $1 = a skills directory -> N_SKL N_CMD
+  N_SKL=0; N_CMD=0; local f l k e d
+  for f in "$1"/*/SKILL.md; do
+    [ -f "$f" ] || continue; k=0; e=0; d=0
+    while IFS= read -r l || [ -n "$l" ]; do
+      l="${l%$'\r'}"
+      case "$l" in ---) d=$((d+1)); [ "$d" -ge 2 ] && break ;; "  kind: command") k=1 ;; "  experimental: true") e=1 ;; esac
+    done < "$f"
+    [ "$e" = 1 ] && continue
+    if [ "$k" = 1 ]; then N_CMD=$((N_CMD+1)); else N_SKL=$((N_SKL+1)); fi
+  done
+}
+count_showcase "$SRC/skills"
 
 h1 '[2/2] Summary · see what will be installed before you confirm'
 echo
 _mt 'full install'; _a="$_M"; _mt '— backend + web + mobile (RN/Expo), every agent and skill'
 row 'Scope' "${B}${_a}${D} ${_M}${R}"
-_mt '%s agents · %s skills will be installed' "${MG}${B}${N_AG}${R}" "${MG}${B}${N_SK}${R}"
+_mt '%s agents · %s skills · %s commands will be installed' "${MG}${B}${N_AG}${R}" "${MG}${B}${N_SKL}${R}" "${MG}${B}${N_CMD}${R}"
 row 'Included'  "$_M"
 _mt 'stack-agnostic — the stack comes from CLAUDE.md ## Stack or the repo (backend-architecture)'
 row 'Backend pattern' "$_M"
@@ -607,7 +622,8 @@ cp -R "$SRC/eval/."     .claude/eval/ 2>/dev/null || true
 # outside the payload, because kit/ ships whole and 104 KB of test code
 # would travel through all four channels only to be removed on arrival.
 cp -R "$SRC/studio/."   .claude/studio/ 2>/dev/null || true
-{ _mt "%s agents, %s skills installed." "$(ls .claude/agents/*.md 2>/dev/null | wc -l | tr -d ' ')" "$(ls -d .claude/skills/*/ 2>/dev/null | wc -l | tr -d ' ')"; echo "  ${_M}"; }
+count_showcase .claude/skills
+{ _mt "%s agents, %s skills and %s commands installed." "$(ls .claude/agents/*.md 2>/dev/null | wc -l | tr -d ' ')" "$N_SKL" "$N_CMD"; echo "  ${_M}"; }
 [ -f "$SRC/settings.json" ] && cp "$SRC/settings.json" .claude/settings.json
 [ -f "$HERE/VERSION" ] && cp "$HERE/VERSION" .claude/VERSION   # make the kit version trackable in the installed project
 # Glob form so every shipped hook/eval is made executable — including ones added later (guard-write.sh,
@@ -721,7 +737,7 @@ else
 fi
 rm -rf "$SRC"
 echo
-{ _mt 'Done. ./.claude + ./CLAUDE.md ready (full install); kit/ deleted.'; echo "== ${_M} =="; }
+{ _mt 'Done. ./.claude + ./CLAUDE.md ready (full install); the installer files are removed.'; echo "== ${_M} =="; }
 { _mt 'Next: 1) fill in the CLAUDE.md project section  2) open Claude Code at the repo root'; echo "${_M}"; }
 { _mt 'Note: if Claude Code is ALREADY running here, restart it — CLAUDE.md and the discipline load at session start.'; echo "${_M}"; }
 { _mt "Tip:  open Claude Code and run /crew-doctor — it checks the install is wired (hooks executable, core.hooksPath set, discipline imported) and scores the project's readiness. CLAUDE.md loads the discipline every session."; echo "${_M}"; }