v1.93.11.0 fix(capy): installer picks the loadout from the owner's agent_model when no model is named (GST-982) (#3872)

* fix(capy): installer picks the loadout from the owner's agent_model preference when no model is named (GST-982)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(capy): label config-sourced agent_model, note the assumption, skip it inside eval runs (GST-982 review)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* docs(loadouts): the Capy preference is an assumption; non-Opus sessions set GSTACK_MODEL (GST-982 review)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(capy): an unresolvable agent_model prints LOADOUT_FALLBACK (GST-982 review)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* test(capy): type the eval env rows for the test typecheck ratchet (GST-982)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

---------

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## [1.93.11.0] - 2026-11-02
+
+**Capy sessions get their model's loadout again without a sentence in the skill (GST-982).** The browse wave (#3858) removed the Capy skill's instruction to run the installer with `GSTACK_MODEL`, to keep `capy/skill/gstack/SKILL.md` under its context budget. After that, every Capy session installed the full skills, including Opus 5.5 sessions that should get `lite`. Now the installer finds the model by itself.
+
+### Fixed
+- **`gstack-capy-install` reads the owner's model preference.** When it gets no `--model` and no `GSTACK_MODEL`, it uses the `agent_model` in the session's `GSTACK_PREFERENCES_FILE` (the Drive skill's `preferences.yaml`, which `gstack-capy-session` already exports), resolved by `gstack-models resolve --role agent`. So `opus` means `claude-opus-5-5` today. The line reads `LOADOUT: … source=preferences` (`source=config` when this machine's `config.yaml` sets it), followed by a `LOADOUT_ASSUMED:` note that the model came from the preference and not the session. Eval runs (`GSTACK_EVAL_DIR` or `EVALS_RUN_ID` set) skip the preference, and a preference `gstack-models` rejects prints `LOADOUT_FALLBACK` instead of serving full silently. `GSTACK_MODEL` and `--model` still win, and a `host` preference or no file still serves full. The frozen bootstrapper `capy/skill/gstack/scripts/install.sh` and the skill text are unchanged.
+
+### Tests
+- `test/gstack-capy-install.test.ts`: an `opus` preference picks lite with `source=preferences` and its own launch file, `GSTACK_MODEL` overrides the preference, a `config.yaml` preference reads `source=config`, eval runs skip the preference, a rejected preference prints `LOADOUT_FALLBACK`, a `host` preference prints no loadout lines, and the real `gstack-models resolve --role agent` keeps the `agent: <provider>/<model>` first line the installer reads.
+
 ## [1.93.10.0] - 2026-11-02
 
 ### Changed

VERSION

@@ -1 +1 @@
-1.93.10.0
+1.93.11.0

agents-digest/gstack-AGENTS.md

@@ -1,4 +1,4 @@
-# gstack digest v1.93.10.0 — regenerate/re-copy after upgrading gstack
+# gstack digest v1.93.11.0 — regenerate/re-copy after upgrading gstack
 
 Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed
 for agent hosts without a full skill install. The full skills add workflows,

bin/gstack-capy-install

@@ -42,7 +42,8 @@
 # mode-0600 file beside the verdict; callers and their children `. <file>` it,
 # never re-type it) and the doctor's trailer (`gstack: ok ...` or `gstack: fail ...`;
 # grep that, never pipe through tail).
-# --model <id>, else GSTACK_MODEL (GSTA-161, docs/loadouts.md), names the model this session runs:
+# --model <id>, else GSTACK_MODEL (GSTA-161, docs/loadouts.md), else the agent_model of the session's
+# GSTACK_PREFERENCES_FILE as gstack-models resolve --role agent reads it (GST-982), names the model this session runs:
 # bin/gstack-modules install picks its loadout, publishes a non-full one beside
 # the full Claude install and prints LOADOUT and SKILL_ROOT lines before the
 # launch lines; a non-full loadout gets its own launch file, keyed by loadout
@@ -723,16 +724,29 @@ fi
 [ "$RC" -eq 0 ] || { [ "$PROFILE" -eq 1 ] && rm -rf "${VERDICT%.json}.profile" 2>/dev/null; verdict_kept || rm -f "$LAUNCH_FILE" 2>/dev/null; LAUNCH_FILE=""; }
 record_verdict "$TRAILER" "$([ "$RC" -eq 0 ] && echo 1 || echo 0)"
 LOADOUT_LINES=""
+if [ "$RC" -eq 0 ] && [ -z "$MODEL" ] && [ -n "${GSTACK_PREFERENCES_FILE:-}" ] && [ -z "${GSTACK_EVAL_DIR:-}${EVALS_RUN_ID:-}" ] && [ "$PROFILE" -eq 0 ] && [ -n "$LAUNCH_FILE" ]; then
+  _ag_out=$(cd "$ROOT" && bin/gstack-models resolve --role agent 2>&1) && _ag_rc=0 || _ag_rc=$?
+  _ag_first=$(printf '%s\n' "$_ag_out" | head -1)
+  case "$_ag_first" in
+    *", this machine") MODEL_SOURCE=config ;;
+    *", the preferences file") MODEL_SOURCE=preferences ;;
+  esac
+  [ "$_ag_rc" -eq 0 ] || LOADOUT_LINES="LOADOUT_FALLBACK: agent_model unreadable: $(printf '%s' "$_ag_first" | tr -d '\000-\037' | cut -c1-200); serving full (docs/troubleshooting.md#loadout-fallback)"
+  [ "$MODEL_SOURCE" = GSTACK_MODEL ] || MODEL=$(printf '%s' "$_ag_first" | sed -n 's|^agent: [a-z0-9-]*/\([A-Za-z0-9._-]*\) .*|\1|p')
+fi
 if [ "$RC" -eq 0 ] && [ -n "$MODEL" ] && [ "$PROFILE" -eq 0 ] && [ -n "$LAUNCH_FILE" ]; then
   _lo_full=$(. "$SELF/bin/gstack-claude-root.sh" && dirname "$(gstack_canonical_claude_root)")
-  if [ "$MODEL_SOURCE" = model-flag ]; then
+  if [ "$MODEL_SOURCE" = GSTACK_MODEL ]; then
-    _lo_out=$(cd "$ROOT" && bin/gstack-modules install --model "$MODEL" --full-root "$_lo_full" 2>&1)
-  else
     _lo_out=$(cd "$ROOT" && GSTACK_MODEL=$MODEL bin/gstack-modules install --full-root "$_lo_full" 2>&1)
+  else
+    _lo_out=$(cd "$ROOT" && bin/gstack-modules install --model "$MODEL" --full-root "$_lo_full" 2>&1)
   fi
   _lo_line=$(printf '%s\n' "$_lo_out" | grep -m1 '^LOADOUT: ')
+  case "$MODEL_SOURCE" in preferences|config) _lo_line=$(printf '%s' "$_lo_line" | sed "s/ source=model-flag/ source=$MODEL_SOURCE/") ;; esac
   _lo_root=$(printf '%s\n' "$_lo_out" | sed -n 's/^SKILL_ROOT: //p' | head -1)
   _lo_notes=$(printf '%s\n' "$_lo_out" | grep -E '^(LOADOUT_INVALID_MODEL|LOADOUT_FALLBACK): ')
+  case "$MODEL_SOURCE" in preferences|config) _lo_notes="${_lo_notes:+$_lo_notes
+}LOADOUT_ASSUMED: model=$MODEL is the owner's agent_model ($MODEL_SOURCE), not read from the session; a session on another model sets GSTACK_MODEL=<its id>" ;; esac
   _lo_name=$(printf '%s' "$_lo_line" | sed -n 's/^LOADOUT: \([a-z0-9-]*\) .*/\1/p')
   _lo_digest=$(printf '%s' "$_lo_line" | sed -n 's/.* digest=\([0-9a-f]\{12\}\) .*/\1/p')
   if [ -n "$_lo_name" ] && [ "$_lo_name" != full ]; then

docs/capy-install.md

@@ -472,8 +472,9 @@ make <dir> writable, or set GSTACK_STATE_ROOT to a writable directory
 (LAUNCH_FILE_UNWRITABLE)`, and a failing run prints no `GSTACK_LAUNCH_FILE` line
 and removes the file an earlier passing run left beside the verdict it rewrites
 (a profile check that keeps a fuller ok verdict leaves that verdict's file).
-With `--model <id>`, or `GSTACK_MODEL=<id>` in its environment (what the Capy
+With `--model <id>`, `GSTACK_MODEL=<id>` in its environment, or else an
-skill passes), an ok run other than `--for review-voice` also prints `LOADOUT: …`
+`agent_model` in the session's `GSTACK_PREFERENCES_FILE` outside an eval run (GST-982, with a
+`LOADOUT_ASSUMED:` note), an ok run other than `--for review-voice` also prints `LOADOUT: …`
 and `SKILL_ROOT: <dir>` just before `GSTACK_LAUNCH`. For a loadout other than
 `full`, `GSTACK_LAUNCH_FILE` then names `<verdict file without .json>.loadout-<name>-<digest>.env`,
 a copy of the verdict's launch file that adds `GSTACK_LOADOUT_SKILL_ROOT`; the

docs/loadouts.md

@@ -44,17 +44,18 @@ bin/gstack-modules install --full-root "$HOME/.claude/skills" --model claude-opu
 
 ### How gstack picks a loadout
 
-gstack never guesses the model. It uses, in order:
+gstack does not detect the running model. It uses, in order:
 
 1. An explicit loadout: `./setup --loadout <name>` on Claude Code, or `bin/gstack-modules install --loadout <name>`. It is served as asked and bound to no model.
 2. An explicit model id: `--model <id>` on the Capy installer (`scripts/install.sh --model <id>`, which passes it to `bin/gstack-capy-install`), else `GSTACK_MODEL`, which the installer also reads.
-3. Nothing: `full`.
+3. On Capy only, with neither and outside an eval run: the owner's `agent_model` (this machine's `config.yaml`, else the Drive skill's `preferences.yaml`), as `gstack-models resolve --role agent` resolves it (`opus` is the newest Opus). This is an assumption, not a detection, and the installer prints `LOADOUT_ASSUMED:` beside it. `host` or no preference means no model.
+4. Nothing: `full`.
 
-A model id is lowercased, its `provider/` prefix and a trailing `[…]` are dropped, and it must then be a plain id (`^[a-z0-9][a-z0-9._-]{0,79}$`). An unmapped id serves `full`; an id that is not plain also serves `full` and prints `LOADOUT_INVALID_MODEL`. Nothing reads Claude Code settings or an inherited loadout variable, so a Sonnet subagent of an Opus session gets `full` unless its own install names Opus.
+A model id is lowercased, its `provider/` prefix and a trailing `[…]` are dropped, and it must then be a plain id (`^[a-z0-9][a-z0-9._-]{0,79}$`). An unmapped id serves `full`; an id that is not plain also serves `full` and prints `LOADOUT_INVALID_MODEL`. Nothing reads Claude Code settings or an inherited loadout variable, so on Claude Code a Sonnet subagent of an Opus session gets `full` unless its own install names Opus. On Capy every session shares the owner's preference, so a session on another model sets `GSTACK_MODEL=<its id>` to get its own loadout.
 
 ### Capy
 
-The Capy skill passes the thread's model id in the environment: `GSTACK_MODEL=<id> bash scripts/install.sh --check --for <skill>`. The variable, unlike `--model`, is safe on a checkout whose installer predates loadouts: that installer ignores it instead of refusing an unknown argument. The installer prints two lines before the launch lines:
+A plain `bash scripts/install.sh --check --for <skill>` picks the loadout from the owner's `agent_model` preference, so the skill needs no model sentence (`source=preferences` on the line below, or `source=config` when this machine's `config.yaml` sets `agent_model`). The preference names the model Garry starts sessions on, not the model reading the skill, so the installer says so in a `LOADOUT_ASSUMED:` line, and it skips the preference inside an eval run (`GSTACK_EVAL_DIR` or `EVALS_RUN_ID` set), where the model under test may differ. A session on another model names its own with `GSTACK_MODEL=<id>`, which wins over the preference and, unlike `--model`, is safe on a checkout whose installer predates loadouts: that installer ignores it instead of refusing an unknown argument. The installer prints two lines before the launch lines:
 
 ```
 LOADOUT: lite model=claude-opus-5-5 source=GSTACK_MODEL digest=3f2a9c01d4e7 reused=yes

package.json

@@ -1,6 +1,6 @@
 {
   "name": "gstack",
-  "version": "1.93.10",
+  "version": "1.93.11",
   "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
   "license": "MIT",
   "type": "module",

test/gstack-capy-install.test.ts

@@ -593,6 +593,82 @@ describe('gstack-capy-install: launch PATH, Bun reuse, CI pins, setup skip, verd
     expect(fs.existsSync(lite)).toBe(true);
   });
 
+  test('with no --model and no GSTACK_MODEL, the agent_model in the session preferences file picks the loadout (GST-982); GSTACK_MODEL still wins and a host preference serves full', () => {
+    const f = fixture();
+    script(path.join(f.root, 'bin/gstack-modules'), [
+      'echo "modules $*" >> "$CAPY_LOG"',
+      'model="${GSTACK_MODEL:-}" full=""; while [ $# -gt 0 ]; do case "$1" in --model) model="$2"; shift ;; --full-root) full="$2"; shift ;; esac; shift; done',
+      'case "$model" in',
+      '  claude-opus-5-5) mkdir -p "$HOME/lite-0123456789ab"; echo "LOADOUT: lite model=claude-opus-5-5 source=model-flag digest=0123456789ab reused=no"; echo "SKILL_ROOT: $HOME/lite-0123456789ab" ;;',
+      '  *) echo "LOADOUT: full model=none source=GSTACK_MODEL reason=unmapped"; echo "SKILL_ROOT: $full" ;;',
+      'esac',
+    ]);
+    script(path.join(f.root, 'bin/gstack-models'), [
+      'echo "models $* prefs=${GSTACK_PREFERENCES_FILE:-}" >> "$CAPY_LOG"',
+      '[ -z "${FAKE_BAD_MODEL:-}" ] || { echo "agent: agent_model $FAKE_BAD_MODEL is not a model"; exit 1; }',
+      '[ -z "${FAKE_CONFIG_MODEL:-}" ] || { echo "agent: anthropic/$FAKE_CONFIG_MODEL (opus) - from agent_model in $HOME/.gstack/config.yaml, this machine"; exit 0; }',
+      'case "$(sed -n "s/^agent_model: *//p" "${GSTACK_PREFERENCES_FILE:-/dev/null}" 2>/dev/null)" in',
+      '  opus) echo "agent: anthropic/claude-opus-5-5 (opus) - from agent_model in $GSTACK_PREFERENCES_FILE, the preferences file" ;;',
+      '  *) echo "agent: host - keep the running agent\'s own model and omit the model field (no preference set)" ;;',
+      'esac',
+    ]);
+    const prefs = path.join(f.home, 'drive-skill/preferences.yaml');
+    write(prefs, 'agent_model: opus\n');
+    const launchOf = (out: string) => out.match(/^GSTACK_LAUNCH_FILE: (.+)$/m)![1];
+    const opus = run(f, ['--for', 'ship'], { GSTACK_PREFERENCES_FILE: prefs });
+    expect(opus.status, opus.stdout + opus.stderr).toBe(0);
+    expect(f.calls().filter(l => l.startsWith('modules '))).toEqual([`modules install --model claude-opus-5-5 --full-root ${f.home}/.claude/skills`]);
+    expect(opus.stdout).toContain(`LOADOUT: lite model=claude-opus-5-5 source=preferences digest=0123456789ab reused=no\nSKILL_ROOT: ${f.home}/lite-0123456789ab\n`);
+    expect(launchOf(opus.stdout)).toBe(verdicts(f)[0].replace(/\.json$/, '.loadout-lite-0123456789ab.env'));
+    expect(opus.stdout).toContain("LOADOUT_ASSUMED: model=claude-opus-5-5 is the owner's agent_model (preferences), not read from the session; a session on another model sets GSTACK_MODEL=<its id>\n");
+
+    const config = run(f, ['--for', 'ship'], { GSTACK_PREFERENCES_FILE: prefs, FAKE_CONFIG_MODEL: 'claude-opus-5-5' });
+    expect(config.status, config.stdout + config.stderr).toBe(0);
+    expect(config.stdout).toContain('LOADOUT: lite model=claude-opus-5-5 source=config digest=0123456789ab');
+    expect(config.stdout).toContain("is the owner's agent_model (config)");
+
+    const modulesBefore = f.calls().filter(l => l.startsWith('modules ')).length;
+    for (const evalEnv of [{ GSTACK_EVAL_DIR: path.join(f.home, 'evals') }, { EVALS_RUN_ID: 'r1' }] as Record<string, string>[]) {
+      const ev = run(f, ['--for', 'ship'], { GSTACK_PREFERENCES_FILE: prefs, ...evalEnv });
+      expect(ev.status, ev.stdout + ev.stderr).toBe(0);
+      expect(ev.stdout).not.toMatch(/^(LOADOUT|LOADOUT_ASSUMED|SKILL_ROOT):/m);
+    }
+    expect(f.calls().filter(l => l.startsWith('modules ')).length).toBe(modulesBefore);
+
+    const bad = run(f, ['--for', 'ship'], { GSTACK_PREFERENCES_FILE: prefs, FAKE_BAD_MODEL: 'gpt-6.1-sol' });
+    expect(bad.status, bad.stdout + bad.stderr).toBe(0);
+    expect(bad.stdout).toContain('LOADOUT_FALLBACK: agent_model unreadable: agent: agent_model gpt-6.1-sol is not a model; serving full (docs/troubleshooting.md#loadout-fallback)\n');
+    expect(bad.stdout).not.toMatch(/^(LOADOUT|SKILL_ROOT):/m);
+    expect(launchOf(bad.stdout)).toBe(verdicts(f)[0].replace(/\.json$/, '.env'));
+    expect(f.calls().filter(l => l.startsWith('modules ')).length).toBe(modulesBefore);
+
+    const env = run(f, ['--for', 'ship'], { GSTACK_PREFERENCES_FILE: prefs, GSTACK_MODEL: 'claude-sonnet-5-5' });
+    expect(env.status, env.stdout + env.stderr).toBe(0);
+    expect(f.calls().filter(l => l.startsWith('modules ')).at(-1)).toBe(`modules install --full-root ${f.home}/.claude/skills`);
+    expect(env.stdout).toContain('LOADOUT: full model=none source=GSTACK_MODEL reason=unmapped\n');
+
+    write(prefs, 'agent_model: host\n');
+    const host = run(f, ['--for', 'ship'], { GSTACK_PREFERENCES_FILE: prefs });
+    expect(host.status, host.stdout + host.stderr).toBe(0);
+    expect(host.stdout).not.toMatch(/^(LOADOUT|SKILL_ROOT):/m);
+    expect(launchOf(host.stdout)).toBe(verdicts(f)[0].replace(/\.json$/, '.env'));
+  });
+
+  test('the first line of gstack-models resolve --role agent keeps the provider/model shape the installer reads (GST-982)', () => {
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'gst982-'));
+    try {
+      for (const [pref, want] of [['opus', /^agent: anthropic\/claude-opus-[0-9a-z.-]+ \(opus\) /], ['host', /^agent: host /]] as const) {
+        const file = path.join(dir, `${pref}.yaml`);
+        fs.writeFileSync(file, `agent_model: ${pref}\n`);
+        const out = spawnSync(path.join(REPO, 'bin/gstack-models'), ['resolve', '--role', 'agent'], { encoding: 'utf8', timeout: 30_000, env: { ...process.env, GSTACK_PREFERENCES_FILE: file, GSTACK_STATE_ROOT: dir, GSTACK_HOME: dir } });
+        expect(out.status, out.stderr).toBe(0);
+        expect(out.stdout.split('\n')[0]).toMatch(want);
+      }
+    } finally {
+      fs.rmSync(dir, { recursive: true, force: true });
+    }
+  });
+
   test('the verdict file records checkout, HEAD, workflow, trailer and the Bun resolution, keyed by checkout and project; --check certifies without setup or downloads', () => {
     const f = fixture();
     const r = run(f, ['--for', 'autoplan']);