agent-bios 0.3.0 → 0.4.0

package/scripts/check-parity.sh

@@ -113,8 +113,39 @@ real Microsoft Excel engine|claude/CLAUDE.md|claude/guides/coding-staged-workflo
 severity contract|README.md|claude/guides/coding-staged-workflow.md
 Ambient state|claude/CLAUDE.md|claude/guides/tooling-gotchas.md
 the full lifecycle of what you create|claude/CLAUDE.md|claude/guides/tooling-gotchas.md
+dual-provider frontier design drafts|claude/CLAUDE.md|claude/guides/cli-multi-model-workflow.md
 ANCHORS
 
+# Domain-manifest gate: bullet<->anchor bijection, file coverage, router
+# co-packaging (config/domains.json vs the monolith); its --self-test proves
+# every negative control still fails, so a green gate is falsifiable.
+if [ -f config/domains.json ]; then
+  python3 scripts/check-domains.py >/dev/null \
+    || { echo "FAIL: domains manifest gate (run scripts/check-domains.py)"; fail=1; }
+  python3 scripts/check-domains.py --self-test >/dev/null \
+    || { echo "FAIL: domains gate self-test missed a negative control"; fail=1; }
+  bash scripts/test-assemble.sh >/dev/null \
+    || { echo "FAIL: assembler scenario suite (run scripts/test-assemble.sh)"; fail=1; }
+fi
+
+# Learning record gate: config/learning.schema.json (collection-loop
+# SSOT) vs its fixtures; --self-test proves every mutation is still caught.
+if [ -f config/learning.schema.json ]; then
+  python3 scripts/check-learning.py >/dev/null \
+    || { echo "FAIL: learning record gate (run scripts/check-learning.py)"; fail=1; }
+  python3 scripts/check-learning.py --self-test >/dev/null \
+    || { echo "FAIL: learning gate self-test missed a negative control"; fail=1; }
+fi
+
+# Lexicon gate: forbid deprecated terminology tokens in live files
+# (LEXICON.md is the SSOT); --self-test proves the detector can fire.
+if [ -f LEXICON.md ]; then
+  python3 scripts/check-lexicon.py >/dev/null \
+    || { echo "FAIL: lexicon gate (run scripts/check-lexicon.py)"; fail=1; }
+  python3 scripts/check-lexicon.py --self-test >/dev/null \
+    || { echo "FAIL: lexicon gate self-test failed"; fail=1; }
+fi
+
 # The launcher's Textual preflight UI tests need the managed venv (textual).
 # Provision it if missing; every non-UI check above runs under system python.
 VENV="${AGENT_LAUNCH_VENV:-$HOME/.local/share/agent-launch/venv}"
@@ -551,7 +582,9 @@ if launcher.is_file() and launch_profile:
             def set_plan(self, plan):
                 self.plan = plan
 
-            def choose(self, title, options, default, allow_back, preview=None, learning_lines=None):
+            def choose(self, title, options, default, allow_back, preview=None, corpus_lines=None):
+                if title == "Mode":
+                    return "builder"
                 if title == "Preset":
                     if preview is not None:
                         balanced = launcher_module.setup_summary_lines(preview("balanced"))
@@ -772,12 +805,17 @@ if launcher.is_file() and launch_profile:
                 os.close(master)
             return bytes(transcript), picker_status
 
+        # Root menu is now a Mode picker (General user / Builder / Session
+        # distill); Builder is the default highlight and lists Custom. Reaching
+        # Custom means: Enter the root to open Builder's preset submenu, then
+        # step down through its fixed presets (Balanced default-highlighted).
         open_custom_steps = (
+            (b"Esc cancel | q cancel", b"\r"),
             (b"native multi-perspective review", b""),
-            (b"Esc cancel | q cancel", b"\x1b[B"),
+            (b"Esc back | q cancel", b"\x1b[B"),
             (b"hybrid onto", b"\x1b[B"),
             (b"high-volume", b"\x1b[B"),
-            (b"learning-refinement", b"\x1b[B"),
+            (b"the standing spawn policy is lifted", b"\x1b[B"),
             (b"Open a settings hub", b"\r"),
         )
         down = b"\x1b[B"
@@ -798,7 +836,7 @@ if launcher.is_file() and launch_profile:
             b"WORKHORSE",
             b"SWEEP",
             b"About highlighted option",
-            b"HELM default for everyday work",
+            b"Tune tiers, review routes, and permissions",
             b"Options (",
         )
         positions = {token: transcript.find(token) for token in layout_tokens}
@@ -809,7 +847,7 @@ if launcher.is_file() and launch_profile:
             positions[b"Current setup"]
             < min(positions[token] for token in (b"FRONTIER", b"HELM", b"WORKHORSE", b"SWEEP"))
             < positions[b"About highlighted option"]
-            < positions[b"HELM default for everyday work"]
+            < positions[b"Tune tiers, review routes, and permissions"]
             < positions[b"Options ("]
         ):
             mark_fail("agent-launch layout is not setup then description then options")
@@ -821,12 +859,27 @@ if launcher.is_file() and launch_profile:
         if picker_status != 130:
             mark_fail(f"agent-launch root Esc returned {picker_status}, want 130")
 
+        # Root Mode picker navigation: Esc from the Builder preset submenu
+        # returns to a freshly re-rendered Mode picker (Builder's own
+        # description reappears) rather than cancelling the launcher outright.
+        _, picker_status = run_picker_scenario(
+            "mode picker navigation",
+            (
+                (b"Esc cancel | q cancel", b"\r"),
+                (b"native multi-perspective review", b"\x1b"),
+                (b"Tune tiers, review routes, and permissions", b"q"),
+            ),
+        )
+        if picker_status != 130:
+            mark_fail(f"agent-launch mode picker Esc-back returned {picker_status}, want 130")
+
         if argv_log.exists():
             argv_log.unlink()
         _, picker_status = run_picker_scenario(
             "launch confirmation q cancellation",
             (
                 (b"Esc cancel | q cancel", b"\r"),
+                (b"HELM default for everyday work", b"\r"),
                 (b"Launch? [Y/n/q]:", b"q\r"),
             ),
             dry_run=False,
@@ -1128,10 +1181,19 @@ if launcher.is_file() and launch_profile:
         # TERM=dumb routes to numbered prompts (textual renders on any usable
         # terminal, so the numbered fallback is gated on TERM/non-TTY/textual
         # availability, not a terminfo probe). 'b' is the numbered back command.
-        custom_number = str(len(fake_data["presets"]) + 1).encode()
+        # The root Mode menu is fixed (General=1, Builder=2, Session distill=3)
+        # regardless of preset count; only Custom's position within Builder's
+        # own preset submenu depends on how many builder-mode presets exist.
+        builder_preset_count = sum(
+            1
+            for data in fake_data["presets"].values()
+            if data.get("mode", "builder") == "builder"
+        )
+        custom_number = str(builder_preset_count + 1).encode()
         transcript, picker_status = run_picker_scenario(
             "numbered fallback (TERM=dumb)",
             (
+                (b"Tune tiers, review routes, and permissions", b"2\n"),
                 (b"Open a settings hub", custom_number + b"\n"),
                 (b"Custom settings", b"b\n"),
                 (b"HELM default for everyday work", b"1\n"),
@@ -1145,27 +1207,27 @@ if launcher.is_file() and launch_profile:
         ):
             mark_fail("agent-launch numbered fallback (TERM=dumb) did not preserve numbered back")
 
-        # Session Learning hub: enter, render status, back out, launch balanced.
-        learning_number = str(len(fake_data["presets"]) + 2).encode()
+        # Session Distill hub: enter, render status, back out, launch balanced.
         transcript, picker_status = run_picker_scenario(
-            "numbered learning hub (TERM=dumb)",
+            "numbered distill hub (TERM=dumb)",
             (
-                (b"Session Learning", learning_number + b"\n"),
+                (b"Session distill", b"3\n"),
                 (b"Versions & rollback", b"4\n"),
+                (b"Tune tiers, review routes, and permissions", b"2\n"),
                 (b"HELM default for everyday work", b"1\n"),
             ),
             term="dumb",
         )
-        learning_panel_rendered = (
+        corpus_panel_rendered = (
             b"Applied version" in transcript or b"not projected yet" in transcript
         )
         if (
             picker_status != 0
             or b"Traceback" in transcript
-            or not learning_panel_rendered
+            or not corpus_panel_rendered
             or b"Preset         Balanced" not in transcript
         ):
-            mark_fail("agent-launch numbered learning hub did not render or return")
+            mark_fail("agent-launch numbered distill hub did not render or return")
 
         passed = invoke([
             sys.executable, str(launcher), "--no-tui", "codex", "--", "exec", "--json", "probe"
@@ -1492,6 +1554,29 @@ if launcher.is_file() and launch_profile:
                         "agent-launch Claude expert overrides were not appended last"
                     )
 
+        # Vanilla (mode=general): the raw backend with nothing applied — no
+        # launch contract, no agents, no permission-bypass flag.
+        vanilla_configured = invoke([
+            sys.executable, str(launcher), "--preset", "vanilla", "--yes", "--dry-run",
+            "claude",
+        ], env=env)
+        if vanilla_configured.returncode != 0:
+            mark_fail(
+                f"agent-launch vanilla dry-run failed: {vanilla_configured.stderr.strip()}"
+            )
+        else:
+            vanilla_argv = json.loads(vanilla_configured.stdout.splitlines()[-1])
+            for forbidden in (
+                "--append-system-prompt", "--agents", "--dangerously-skip-permissions",
+            ):
+                if forbidden in vanilla_argv:
+                    mark_fail(f"agent-launch vanilla preset unexpectedly carried {forbidden}")
+            if vanilla_argv != [str(backend)]:
+                mark_fail(
+                    "agent-launch vanilla preset must project the bare backend with "
+                    f"nothing applied: {vanilla_argv!r}"
+                )
+
         # Cross-family default: each main routes every review route to the OPPOSITE
         # model family, with concrete tools/paths/bindings named in the contract.
         # The cross CODEX_HOME needs the reviewer wrappers (bin) and the same-family
@@ -1766,5 +1851,5 @@ then
   fail=1
 fi
 
-[ "$fail" -eq 0 ] && echo "PARITY OK: mirrors, globals, guides, launch profile, bypass paths, role bindings, and wrapper defaults aligned"
+[ "$fail" -eq 0 ] && echo "PARITY OK: mirrors, globals, guides, domain manifest, assembler, launch profile, bypass paths, role bindings, and wrapper defaults aligned"
 exit "$fail"

package/scripts/install.sh

@@ -102,6 +102,29 @@ deploy_glob() {
   done
 }
 
+# ---- packaged (domain-selected) corpus deploy ----------------------------
+# Active when --domains is passed or a prior selection exists in STATE_DIR.
+# assemble.py owns the corpus surfaces (central tree, entry seeding, codex
+# marker region, settings merge); the entry CLAUDE.md and AGENTS.md are NOT
+# manifested — the entry is user-owned after seeding, AGENTS.md holds a
+# personal region — so uninstall never deletes them.
+packaged_mode() { [ "${DOMAINS_SET:-0}" = 1 ] || [ -f "$STATE_DIR/selection.json" ]; }
+
+assemble_packaged() {
+  local args=(--claude-dir "$CLAUDE_DIR" --codex-dir "$CODEX_DIR" --state-dir "$STATE_DIR") rc=0
+  [ "${DOMAINS_SET:-0}" = 1 ] && args+=(--domains "$DOMAINS_ARG")
+  [ "$DRY_RUN" = 1 ] && args+=(--dry-run)
+  python3 "$REPO/scripts/assemble.py" "${args[@]}" || rc=$?
+  if [ "$rc" = 2 ]; then
+    log "packaged: entry file needs user action (import line missing); central content will not load until it is added"
+  elif [ "$rc" != 0 ]; then
+    return 1
+  fi
+  if [ "$DRY_RUN" != 1 ]; then
+    find "$CLAUDE_DIR/central" "$CODEX_DIR/guides" -type f 2>/dev/null >> "$MANIFEST"
+  fi
+}
+
 add_zsh_hook() {
   if [ -f "$ZSHRC" ] && grep -qF "$HOOK_MARK" "$ZSHRC"; then
     info "zsh hook present  $ZSHRC"
@@ -399,12 +422,17 @@ cmd_install() {
     : > "$MANIFEST"
   fi
   log "Deploying agent-bios from $REPO"
-  deploy_file "$REPO/claude/CLAUDE.md" "$CLAUDE_DIR/CLAUDE.md"
-  deploy_glob "$REPO/claude/guides" "*.md" "$CLAUDE_DIR/guides"
-  deploy_glob "$REPO/claude/agents" "*.md" "$CLAUDE_DIR/agents"
-  deploy_glob "$REPO/claude/hooks" "*.py" "$CLAUDE_DIR/hooks" "+x"
-  deploy_file "$REPO/codex/AGENTS.md" "$CODEX_DIR/AGENTS.md"
-  deploy_glob "$REPO/codex/guides" "*.md" "$CODEX_DIR/guides"
+  if packaged_mode; then
+    log "packaged mode: assembling selected domains (state: $STATE_DIR/selection.json)"
+    assemble_packaged || exit 1
+  else
+    deploy_file "$REPO/claude/CLAUDE.md" "$CLAUDE_DIR/CLAUDE.md"
+    deploy_glob "$REPO/claude/guides" "*.md" "$CLAUDE_DIR/guides"
+    deploy_glob "$REPO/claude/agents" "*.md" "$CLAUDE_DIR/agents"
+    deploy_glob "$REPO/claude/hooks" "*.py" "$CLAUDE_DIR/hooks" "+x"
+    deploy_file "$REPO/codex/AGENTS.md" "$CODEX_DIR/AGENTS.md"
+    deploy_glob "$REPO/codex/guides" "*.md" "$CODEX_DIR/guides"
+  fi
   deploy_glob "$REPO/codex/agents" "*.toml" "$CODEX_DIR/agents"
   codex_config_additions merge || exit 1
   deploy_file "$REPO/scripts/codex-run.sh"  "$CODEX_DIR/bin/codex-run"  "+x"
@@ -424,10 +452,10 @@ cmd_install() {
       || log "warning: venv provisioning failed (numbered-prompt fallback applies)"
   fi
   add_zsh_hook
-  if python3 "$REPO/scripts/session-learning/learning-state.py" project --repo "$REPO" >/dev/null 2>&1; then
-    info "learning-status projected"
+  if python3 "$REPO/scripts/session-distill/corpus-state.py" project --repo "$REPO" >/dev/null 2>&1; then
+    info "corpus-status projected"
   else
-    log "note: learning-status projection unavailable (versions.json/ledger missing?)"
+    log "note: corpus-status projection unavailable (versions.json/ledger missing?)"
   fi
   log ""
   log "Verifying deployment..."
@@ -448,15 +476,32 @@ verify_present() { if [ -f "$1" ]; then return 0; else log "missing  $1"; return
 
 cmd_verify() {
   local fail=0 gp
-  verify_match "$REPO/claude/CLAUDE.md"        "$CLAUDE_DIR/CLAUDE.md"        || fail=1
-  verify_match "$REPO/codex/AGENTS.md"         "$CODEX_DIR/AGENTS.md"         || fail=1
+  if packaged_mode; then
+    # Packaged: corpus surfaces are selection-derived, not repo-identical.
+    # The entry file is user-owned — READ-check the import line, never rewrite.
+    python3 "$REPO/scripts/check-domains.py" >/dev/null 2>&1 && info "domains gate OK" || { log "domains gate FAILED"; fail=1; }
+    verify_present "$CLAUDE_DIR/central/bundle.md" || fail=1
+    if grep -qF '@central/bundle.md' "$CLAUDE_DIR/CLAUDE.md" 2>/dev/null; then
+      info "entry import line present"
+    else
+      log "entry $CLAUDE_DIR/CLAUDE.md lacks '@central/bundle.md' — central corpus is NOT loading"; fail=1
+    fi
+    if grep -qF 'agent-bios:central:start' "$CODEX_DIR/AGENTS.md" 2>/dev/null; then
+      info "codex central region present"
+    else
+      log "codex AGENTS.md central region missing"; fail=1
+    fi
+  else
+    verify_match "$REPO/claude/CLAUDE.md"        "$CLAUDE_DIR/CLAUDE.md"        || fail=1
+    verify_match "$REPO/codex/AGENTS.md"         "$CODEX_DIR/AGENTS.md"         || fail=1
+    for gp in "$REPO"/claude/guides/*.md; do verify_present "$CLAUDE_DIR/guides/$(basename "$gp")" || fail=1; done
+    for gp in "$REPO"/claude/agents/*.md; do verify_present "$CLAUDE_DIR/agents/$(basename "$gp")" || fail=1; done
+    for gp in "$REPO"/claude/hooks/*.py;  do verify_present "$CLAUDE_DIR/hooks/$(basename "$gp")"  || fail=1; done
+    for gp in "$REPO"/codex/guides/*.md;  do verify_present "$CODEX_DIR/guides/$(basename "$gp")"  || fail=1; done
+  fi
   verify_match "$REPO/scripts/agent-launch.py" "$BIN_DIR/agent-launch"        || fail=1
   verify_match "$REPO/config/agent-launch.toml" "$LAUNCH_DIR/profiles.toml"   || fail=1
   verify_match "$REPO/shell/agent-launch.zsh"  "$LAUNCH_DIR/shell.zsh"        || fail=1
-  for gp in "$REPO"/claude/guides/*.md; do verify_present "$CLAUDE_DIR/guides/$(basename "$gp")" || fail=1; done
-  for gp in "$REPO"/claude/agents/*.md; do verify_present "$CLAUDE_DIR/agents/$(basename "$gp")" || fail=1; done
-  for gp in "$REPO"/claude/hooks/*.py;  do verify_present "$CLAUDE_DIR/hooks/$(basename "$gp")"  || fail=1; done
-  for gp in "$REPO"/codex/guides/*.md;  do verify_present "$CODEX_DIR/guides/$(basename "$gp")"  || fail=1; done
   python3 - "$CODEX_DIR/agents" <<'PY' && info "agent TOMLs OK" || fail=1
 import sys, pathlib, tomllib
 root = pathlib.Path(sys.argv[1])
@@ -526,6 +571,49 @@ cmd_uninstall() {
   log "Kept: managed venv + backups under $STATE_DIR (rm -rf \"$STATE_DIR\" to purge)."
 }
 
+cmd_onboard() {
+  log "agent-bios onboarding — pick your domain packages (core + infra always install)"
+  local names=() line i=1 choice sel="" n picks
+  while IFS= read -r line; do names+=("$line"); done \
+    < <(python3 -c "import json;print('\n'.join(sorted(json.load(open('$REPO/config/domains.json'))['domains'])))")
+  [ "${#names[@]}" -ge 1 ] || { log "no domains found in config/domains.json"; exit 1; }
+  if [ "$DOMAINS_SET" = 1 ]; then
+    # Non-interactive path, per this installer's input contract (stdin is
+    # detached at the top of the script): selection arrives as domain names.
+    [ "$DOMAINS_ARG" = none ] && DOMAINS_ARG=""
+    sel="$DOMAINS_ARG"
+  elif ( : </dev/tty ) 2>/dev/null; then
+    for line in "${names[@]}"; do info "$i) $line"; i=$((i+1)); done
+    printf 'Select by number, comma-separated (empty = core+infra only): '
+    read -r choice </dev/tty || choice=""
+    if [ -n "$choice" ]; then
+      # bash 3.2 + set -u: expanding an EMPTY array errors, so split only
+      # when there is input; empty input means core+infra only.
+      IFS=',' read -ra picks <<<"$choice"
+      for n in "${picks[@]}"; do
+        n="${n// /}"; [ -n "$n" ] || continue
+        case "$n" in (*[!0-9]*) log "invalid selection: $n"; exit 2 ;; esac
+        [ "$n" -ge 1 ] && [ "$n" -le "${#names[@]}" ] || { log "selection out of range: $n"; exit 2; }
+        sel="$sel${sel:+,}${names[$((n-1))]}"
+      done
+    fi
+  else
+    log "onboard needs a terminal or an explicit selection — run:"
+    log "  agent-bios onboard --domains a,b   (or --domains none for core+infra only)"
+    exit 2
+  fi
+  DOMAINS_ARG="$sel"; DOMAINS_SET=1
+  log "selection: ${sel:-<core+infra only>}"
+  cmd_install
+  log ""
+  log "Activation canary (proves the bundle loads in a live session)..."
+  if [ "$DRY_RUN" = 1 ]; then info "[dry-run] skip canary probe"; return; fi
+  bash "$REPO/scripts/canary.sh" || {
+    log "ONBOARDING INCOMPLETE: the bundle is installed but not loading — fix the cause above and re-run: agent-bios verify"
+    exit 1
+  }
+}
+
 cmd_status() {
   local version p
   if [ -d "$REPO/.git" ]; then
@@ -561,6 +649,7 @@ usage() {
 agent-bios — deploy the Claude/Codex instruction SSOT into $HOME (by copy).
 
   agent-bios install     deploy into this environment (backs up + verifies)
+  agent-bios onboard     interactive domain selection + packaged install + activation canary
   agent-bios verify      check the deployed state matches the source
   agent-bios status      show what is installed and where
   agent-bios update      git pull + reinstall (clone), or print the npm update line
@@ -568,6 +657,13 @@ agent-bios — deploy the Claude/Codex instruction SSOT into $HOME (by copy).
   agent-bios help
 
 Flags: --dry-run       print actions without changing anything
+       --domains a,b  packaged mode (install/onboard): assemble ONLY the named
+                      domain packages (plus core+infra) instead of the full
+                      corpus; with onboard, 'none' means core+infra only. The
+                      selection persists in the state dir, so later
+                      installs/updates stay packaged until the selection file
+                      is removed. Default (no flag, no saved selection) keeps
+                      today's full-corpus deploy.
        --with a,b     also install the named optional dependencies (install only).
                       Without it, install offers each missing one when the terminal
                       is interactive, and otherwise just prints its install line.
@@ -581,11 +677,15 @@ EOF
 CMD="${1:-help}"
 if [ $# -gt 0 ]; then shift; fi
 WITH=""
+DOMAINS_ARG=""
+DOMAINS_SET=0
 while [ $# -gt 0 ]; do
   case "$1" in
     --dry-run) DRY_RUN=1 ;;
     --with) shift; WITH="${1:-}"; [ -n "$WITH" ] || { log "--with needs a comma-separated capability list"; exit 2; } ;;
     --with=*) WITH="${1#--with=}"; [ -n "$WITH" ] || { log "--with needs a comma-separated capability list"; exit 2; } ;;
+    --domains) shift; DOMAINS_ARG="${1:-}"; DOMAINS_SET=1; [ -n "$DOMAINS_ARG" ] || { log "--domains needs a comma-separated domain list (use onboard for core-only)"; exit 2; } ;;
+    --domains=*) DOMAINS_ARG="${1#--domains=}"; DOMAINS_SET=1; [ -n "$DOMAINS_ARG" ] || { log "--domains needs a comma-separated domain list (use onboard for core-only)"; exit 2; } ;;
     *) log "unknown flag: $1"; exit 2 ;;
   esac
   shift
@@ -593,6 +693,7 @@ done
 
 case "$CMD" in
   install)   cmd_install ;;
+  onboard)   cmd_onboard ;;
   update)    cmd_update ;;
   uninstall) cmd_uninstall ;;
   verify)    if cmd_verify; then log "VERIFY OK"; else log "VERIFY FAILED"; exit 1; fi ;;