@caddis/cli 0.4.53 → 0.4.54

package/bundles/antigravity-plugin/guard_agy.py

@@ -32,6 +32,7 @@ from __future__ import annotations
 import json
 import os
 import re
+import shlex
 import sys
 
 # Functional identity, mirrored from scripts/claudster_config.py.
@@ -87,6 +88,187 @@ _CATASTROPHIC_TARGET = re.compile(
 _FORK_BOMB = re.compile(r":\(\)\s*\{\s*:\s*\|\s*:\s*&\s*\}\s*;\s*:")
 
 
+# `git add` of everything. A tokenizer, not a regex: three regex rounds each opened a new gap, and one backtracked
+# exponentially. Linear in the command length. Copied from hooks/guard.py; test_guard_agy.py checks they agree.
+_GA_SEPARATORS = {";", "&", "&&", "|", "||", "(", ")", "{", "}", "$", "!", "\n"}
+_GA_PUNCT = set(";&|(){}$!\n")  # a token made only of these is a separator (`$(`, `&&`, `);`, `|!`)
+_GA_KEYWORDS = {"if", "then", "else", "elif", "do", "while", "until"}  # the next word starts a command
+_GA_WRAPPERS = {"sudo", "command", "env", "exec", "nohup", "time", "timeout", "nice", "xargs", "call", "wsl",
+                "busybox"}
+_GA_SHELLS = {"bash", "sh", "zsh", "dash", "pwsh", "powershell", "cmd"}
+_GA_JOINING_SHELLS = {"cmd", "pwsh", "powershell"}  # they run ALL the words after the flag; bash -c runs one
+_GA_SHELL_RUN_FLAGS = {"-c", "/c", "/k", "-command", "-commandwithargs"}
+_GA_GIT_VALUE_OPTS = {"-c", "-C", "--git-dir", "--work-tree", "--namespace", "--super-prefix", "--config-env"}
+_GA_EVERYTHING = {"-A", "--all", ".", "./", ".\\", ":/", "*"}
+
+
+def _ga_sep(token: str) -> bool:
+    """A command separator: one of _GA_SEPARATORS, or a token made only of their characters (`$(`, `|!`)."""
+    return token in _GA_SEPARATORS or (bool(token) and set(token) <= _GA_PUNCT)
+
+
+def _ga_name(token: str) -> str:
+    """`/usr/bin/git`, `"C:\\Program Files\\Git\\cmd\\git.exe"`, `GIT` -> `git`."""
+    base = re.split(r"[\\/]", token)[-1].lower()
+    return base[:-4] if base.endswith(".exe") else base
+
+
+def _ga_tokens(text: str) -> list[str]:
+    """Shell tokens, quotes removed (also mid-token: --work-tree="my dir"), backslashes kept (Windows paths).
+    ; & | ( ) { } $ ! and newlines become separator tokens."""
+    def lex(quotes: str) -> list[str]:
+        lexer = shlex.shlex(text.replace("`", "\n"), posix=True, punctuation_chars=";&|(){}$!\n")
+        lexer.whitespace, lexer.whitespace_split, lexer.escape, lexer.quotes = " \t\r", True, "", quotes
+        return list(lexer)
+    try:
+        return lex("'\"")
+    except ValueError:  # unbalanced quotes: read it again without quote handling
+        return lex("")
+
+
+def git_add_everything(command: str, depth: int = 0) -> bool:
+    """True when `command` runs `git add` with -A/--all (also bundled, e.g. -Av), `.`, `./`, `:/` or `*`.
+
+    `git` must start a command (after a separator, a shell keyword such as `do`, VAR=value, or a wrapper such as
+    sudo, env, xargs, nice -n 10), with any path, case or .exe. `bash -c "..."`, `pwsh -Command ...` and
+    `cmd /c ...` are read inside. Only git's global options may stand between `git` and `add`, so
+    `git commit -m "add ."` and `printf "git add ."` pass. Callers pass heredoc-free text and must treat an
+    exception as "no match": this check must never stop the guard's other rules.
+    """
+    if depth > 3:
+        return False
+    text = re.sub(r"\\\r?\n|`\r?\n", " ", command)  # line continuations (bash \, PowerShell `)
+    tokens = _ga_tokens(text)
+    i, start = 0, True
+    while i < len(tokens):
+        tok = tokens[i]
+        if _ga_sep(tok):
+            i, start = i + 1, True
+            continue
+        if not start:
+            i += 1
+            continue
+        if tok in _GA_KEYWORDS or re.match(r"^\w+=", tok):  # `do git ...`, VAR=value before the command
+            i += 1
+            continue
+        name = _ga_name(tok)
+        if name in _GA_WRAPPERS:  # skip the wrapper's own flags and values up to the real command
+            i += 1
+            while i < len(tokens) and not _ga_sep(tokens[i]) and _ga_name(tokens[i]) not in ({"git"} | _GA_SHELLS):
+                i += 1
+            continue
+        start = False
+        if name in _GA_SHELLS:
+            j = i + 1
+            while j < len(tokens) and not _ga_sep(tokens[j]):
+                if tokens[j].lower() in _GA_SHELL_RUN_FLAGS:
+                    rest = tokens[j + 1:]
+                    end = next((k for k, t in enumerate(rest) if _ga_sep(t)), len(rest))
+                    if end == 0:  # `bash -c` with nothing after it
+                        break
+                    inner = " ".join(rest[:end]) if name in _GA_JOINING_SHELLS else rest[0]
+                    if git_add_everything(inner, depth + 1):
+                        return True
+                    break
+                j += 1
+            i = j
+            continue
+        if name != "git":
+            i += 1
+            continue
+        j = i + 1
+        while j < len(tokens) and tokens[j].startswith("-") and not _ga_sep(tokens[j]):
+            j += 2 if tokens[j] in _GA_GIT_VALUE_OPTS else 1
+        if j >= len(tokens) or tokens[j] != "add":
+            i = j
+            continue
+        after_dashdash = False
+        for arg in tokens[j + 1:]:
+            if _ga_sep(arg):
+                break
+            if arg == "--":
+                after_dashdash = True
+                continue
+            if after_dashdash and arg.startswith("-"):  # `git add -- -A` names a file called -A
+                continue
+            if arg in _GA_EVERYTHING or re.fullmatch(r"-[a-zA-Z]*A[a-zA-Z]*", arg):
+                return True
+        i = j + 1
+    return False
+
+
+# Copied from hooks/guard.py: a heredoc body that only carries data is not a command.
+_HEREDOC_MARKER = re.compile(
+    r"<<(-?)[ \t]*(?:'([A-Za-z_][A-Za-z_0-9]*)'|\"([A-Za-z_][A-Za-z_0-9]*)\"|([A-Za-z_][A-Za-z_0-9]*))"
+)
+_HEREDOC_INTERPRETER = re.compile(
+    r"^\s*(?:\S*/)?(?:bash|sh|zsh|dash|python|python3|py|node|ruby|perl|pwsh|powershell)\b",
+    re.I,
+)
+
+
+def _unquoted_at(line: str, position: int) -> bool:
+    """Whether a token at position is outside single and double quotes on this line."""
+    quote = ""
+    escaped = False
+    for char in line[:position]:
+        if escaped:
+            escaped = False
+        elif char == "\\" and quote != "'":
+            escaped = True
+        elif quote:
+            if char == quote:
+                quote = ""
+        elif char in "'\"":
+            quote = char
+    return not quote and not escaped
+
+
+def _without_heredoc_bodies(command: str) -> str:
+    """Skip complete data-only heredocs; preserve executable and incomplete bodies."""
+    lines = command.splitlines(keepends=True)
+    kept: list[str] = []
+    i = 0
+    while i < len(lines):
+        line = lines[i]
+        markers = []
+        executable = False
+        for marker in _HEREDOC_MARKER.finditer(line):
+            pos = marker.start()
+            if (pos and not (line[pos - 1].isspace() or line[pos - 1] in ";&|")
+                    or not _unquoted_at(line, pos)):
+                continue
+            markers.append((next(group for group in marker.groups()[1:] if group is not None),
+                            marker.group(1) == "-"))
+            owner = re.split(r"[;&|]", line[:pos])[-1]
+            executable |= bool(_HEREDOC_INTERPRETER.match(owner))
+        if not markers:
+            kept.append(line)
+            i += 1
+            continue
+        end = i
+        for tag, strip_tabs in markers:
+            end += 1
+            while end < len(lines):
+                candidate = lines[end].rstrip("\r\n")
+                if (candidate.lstrip("\t") if strip_tabs else candidate) == tag:
+                    break
+                end += 1
+            if end == len(lines):
+                return command
+        kept.extend(lines[i:end + 1] if executable else [line])
+        i = end + 1
+    return "".join(kept)
+
+
+def _git_add_everything_safe(text: str) -> bool:
+    """git_add_everything, but an error is "no match": one rule must never switch off the rest of the guard."""
+    try:
+        return git_add_everything(text)
+    except Exception:  # noqa: BLE001 - fail open for this rule only; the other rules still run
+        return False
+
+
 def _strip_quotes(s: str) -> str:
     """Drop quotes so quoted flags/targets still match (rm '-rf' \"/\" → rm -rf /)."""
     return s.replace('"', "").replace("'", "")
@@ -191,6 +373,8 @@ def classify_bash(command: str) -> tuple[str, str]:
         return "deny", "fork bomb"
     if re.search(r"\bdd\b.*\bof=/dev/", nl) or re.search(r"\bmkfs", nl) or re.search(r">\s*/dev/sd[a-z]", nl):
         return "deny", "writes directly to a disk device"
+    if _git_add_everything_safe(_without_heredoc_bodies(c)):
+        return "deny", "git add of everything (-A, --all, . or :/): stage named files only"
 
     # ── ask: destructive-but-legit ──
     if rm_rf:

package/bundles/antigravity-plugin/plugin.json

@@ -1,5 +1,5 @@
 {
   "name": "caddis",
-  "version": "1.3.139",
+  "version": "1.3.140",
   "description": "caddis — agent-agnostic dev harness as an Antigravity (agy) plugin: TDD skills + commands, single-sourced with the Claude Code plugin."
 }

package/bundles/antigravity-plugin/scripts/caddis_budget.py

@@ -0,0 +1,283 @@
+"""Growth control: size budgets and archive rules for handoff-grown files."""
+
+from __future__ import annotations
+
+import argparse
+from pathlib import Path
+import re
+import sys
+from typing import NamedTuple
+
+try:
+    import caddis_exit
+except ModuleNotFoundError as exc:
+    if exc.name != "caddis_exit":
+        raise
+    sys.path.insert(0, str(Path(__file__).resolve().parent))
+    sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "claude-harness" / "scripts"))
+    import caddis_exit
+
+
+class Budget(NamedTuple):
+    glob: str
+    unit: str
+    limit: int
+    rule: str
+
+
+class Over(NamedTuple):
+    path: str
+    unit: str
+    value: int
+    limit: int
+    rule: str
+
+
+BUDGETS: tuple[Budget, ...] = (
+    Budget(
+        ".caddis/relay/*.md",
+        "chars",
+        4000,
+        "rewrite: current state only, git keeps history",
+    ),
+    Budget(
+        ".caddis/relay/*.tasks.md",
+        "table rows",
+        40,
+        "done rows are never written (Phase 1)",
+    ),
+    Budget(
+        ".caddis/relay.md",
+        "lines",
+        60,
+        "rendered by render-index; drop relays with status closed",
+    ),
+    Budget(
+        ".caddis/open-work.md",
+        "rows with Status done",
+        10,
+        "move done rows to .caddis/done/open-work-archive.md with a one-line pointer",
+    ),
+    Budget(
+        "DECISIONS.md, .caddis/DECISIONS.md, .caddis/decisions/DECISIONS.md",  # the last is the fleet's real path
+        "History table rows",
+        30,
+        "move older rows to DECISIONS-archive.md beside it",
+    ),
+    Budget(
+        ".caddis/usage-log.jsonl",
+        "bytes",
+        5000000,
+        "rotate to usage-log-<YYYYMM>.jsonl",
+    ),
+    Budget(
+        ".caddis/session-state/*.md",
+        "files",
+        8,
+        "session_state.py prune (not in a shared checkout)",
+    ),
+    Budget(
+        ".caddis/kb/*.md",
+        "lines per file",
+        300,
+        "split, or move closed parts to .caddis/kb/done/",
+    ),
+)
+
+
+class _ParsedTable(NamedTuple):
+    heading: str | None
+    columns: list[str]
+    data_rows: list[list[str]]
+
+
+def _split_cells(line: str) -> list[str]:
+    s = line.strip()
+    if not s:
+        return []
+    parts = re.split(r"(?<!\\)\|", s)
+    if parts and parts[0] == "":
+        parts.pop(0)
+    if parts and parts[-1] == "":
+        parts.pop(-1)
+    return [p.strip() for p in parts]
+
+
+def _is_separator(line: str) -> bool:
+    cells = _split_cells(line)
+    if not cells:
+        return False
+    for cell in cells:
+        c = cell.strip()
+        if not c or any(ch not in "-: " for ch in c) or "-" not in c:
+            return False
+    return True
+
+
+def _parse_tables(text: str) -> tuple[list[_ParsedTable], int]:
+    lines = text.splitlines()
+    tables: list[_ParsedTable] = []
+    loose_rows = 0
+    n = len(lines)
+    i = 0
+    while i < n:
+        if lines[i].lstrip().startswith("|"):
+            run_start = i
+            while i < n and lines[i].lstrip().startswith("|"):
+                i += 1
+            run = lines[run_start:i]
+            if len(run) >= 2 and _is_separator(run[1]):
+                heading: str | None = None
+                for k in range(run_start - 1, -1, -1):
+                    s = lines[k].strip()
+                    if s.startswith("#"):
+                        heading = s.lstrip("#").strip()
+                        break
+                columns = _split_cells(run[0])
+                data_rows = [_split_cells(r) for r in run[2:] if not _is_separator(r)]
+                tables.append(_ParsedTable(heading=heading, columns=columns, data_rows=data_rows))
+            else:
+                for r in run:
+                    if not _is_separator(r) and not r.lstrip().startswith(("| Old #", "| # |")):
+                        loose_rows += 1
+        else:
+            i += 1
+    return tables, loose_rows
+
+
+def _count_table_rows(text: str) -> int:
+    tables, loose_rows = _parse_tables(text)
+    return sum(len(t.data_rows) for t in tables) + loose_rows
+
+
+def _norm_status(cell: str) -> str:
+    return cell.replace("**", "").replace("__", "").replace("`", "").strip().lower()
+
+
+def _count_done_rows(text: str) -> int:
+    tables, _ = _parse_tables(text)
+    done_re = re.compile(r"^done\b")
+    count = 0
+    if tables:
+        for t in tables:
+            col_map = {c.lower(): idx for idx, c in enumerate(t.columns)}
+            if "status" in col_map:
+                s_idx = col_map["status"]
+                for row in t.data_rows:
+                    if len(row) > s_idx and done_re.search(_norm_status(row[s_idx])):
+                        count += 1
+            else:
+                for row in t.data_rows:
+                    if any(_norm_status(c) == "done" for c in row):
+                        count += 1
+        return count
+
+    for line in text.splitlines():
+        if line.lstrip().startswith("|") and not _is_separator(line):
+            cells = _split_cells(line)
+            if any(_norm_status(c) == "done" for c in cells):
+                count += 1
+    return count
+
+
+def _count_history_rows(text: str) -> int:
+    tables, loose_rows = _parse_tables(text)
+    history_tables = [t for t in tables if t.heading and "history" in t.heading.lower()]
+    if history_tables:
+        return sum(len(t.data_rows) for t in history_tables)
+    return sum(len(t.data_rows) for t in tables) + loose_rows
+
+
+def _measure_file(path: Path, unit: str) -> int:
+    if unit == "bytes":
+        return path.stat().st_size
+    text = path.read_text(encoding="utf-8", errors="ignore")
+    if unit == "chars":
+        return len(text)
+    if unit == "table rows":
+        return _count_table_rows(text)
+    if unit in ("lines", "lines per file"):
+        return len(text.splitlines())
+    if unit == "rows with Status done":
+        return _count_done_rows(text)
+    if unit == "History table rows":
+        return _count_history_rows(text)
+    raise ValueError(f"unknown budget unit: {unit}")
+
+
+def check(root: Path) -> list[Over]:
+    """Return every file under ``root`` that exceeds its configured size budget."""
+    root = Path(root)
+    overs: list[Over] = []
+
+    for budget in BUDGETS:
+        if budget.unit == "files":
+            files = [p for p in sorted(root.glob(budget.glob)) if p.is_file()]
+            value = len(files)
+            if value > budget.limit:
+                overs.append(Over(budget.glob, budget.unit, value, budget.limit, budget.rule))
+            continue
+
+        patterns = [g.strip() for g in budget.glob.split(",")]
+        for pattern in patterns:
+            if "*" in pattern:
+                candidates = [p for p in sorted(root.glob(pattern)) if p.is_file()]
+            else:
+                candidate = root / pattern
+                candidates = [candidate] if candidate.is_file() else []
+
+            for p in candidates:
+                if pattern == ".caddis/relay/*.md" and p.name.endswith(".tasks.md"):
+                    continue
+                rel = p.relative_to(root).as_posix()
+                value = _measure_file(p, budget.unit)
+                if value > budget.limit:
+                    overs.append(Over(rel, budget.unit, value, budget.limit, budget.rule))
+
+    return overs
+
+
+def main(argv: list[str] | None = None) -> int:
+    if sys.stdout and hasattr(sys.stdout, "reconfigure"):
+        try:
+            sys.stdout.reconfigure(encoding="utf-8")
+        except Exception:
+            pass
+    if sys.stderr and hasattr(sys.stderr, "reconfigure"):
+        try:
+            sys.stderr.reconfigure(encoding="utf-8")
+        except Exception:
+            pass
+
+    parser = argparse.ArgumentParser(
+        prog="caddis_budget.py",
+        description="Check handoff-grown files against their size budgets.",
+    )
+    sub = parser.add_subparsers(dest="cmd", required=True)
+    p_check = sub.add_parser("check", help="Check files under --root against BUDGETS.")
+    p_check.add_argument("--root", type=Path, default=Path("."), help="Repository root.")
+    p_check.add_argument(
+        "--report",
+        action="store_true",
+        help="Report files over budget but always exit 0.",
+    )
+
+    args = parser.parse_args(argv)
+    try:
+        overs = check(args.root)
+    except OSError as exc:  # a locked or unreadable file (Windows): --report still never fails the caller
+        print(f"[BUDGET] could not read every file: {exc}")
+        return caddis_exit.CLEAN if args.report else caddis_exit.NOT_RUN
+    if not overs:
+        print("[BUDGET] all within budget")
+        return caddis_exit.CLEAN
+
+    for over in overs:
+        print(f"OVER {over.path}: {over.value} {over.unit} > {over.limit}; fix: {over.rule}")
+    if args.report:
+        return caddis_exit.CLEAN
+    return caddis_exit.BLOCKED
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

package/bundles/antigravity-plugin/scripts/caddis_checkout_guard.py

@@ -59,7 +59,11 @@ def get_porcelain_lines(repo: Path) -> tuple[int, list[str]]:
     return 0, entries
 
 
-SHARED_RE = re.compile(r"^Shared checkout:\s*yes\b", re.IGNORECASE | re.MULTILINE)
+# A plain line, or a list item and/or bold: `- **Shared checkout: yes** (...)` (watch-sight's form until 16 Oct).
+# After `yes` only the line end or a reason may follow: `(`, a dash, `.`, `:`, `,`, `;` or a `#` comment. A word
+# after `yes` is prose ("Shared checkout: yes means ..."), not a declaration.
+SHARED_RE = re.compile(r"^[ \t]*(?:[-*][ \t]+)?(?:\*\*)?Shared checkout:[ \t]*yes(?:\*\*)?[ \t]*(?:$|[(\-–—.:,;#])",
+                       re.IGNORECASE | re.MULTILINE)
 
 
 def entry_paths(entry: str) -> list[str]: