codebase-index 2.1.1__tar.gz → 2.1.2__tar.gz

codebase_index-2.1.2/.claude/skills/codebase-index/.skill_version

@@ -0,0 +1 @@
+2.1.2

{codebase_index-2.1.1 → codebase_index-2.1.2}/.claude-plugin/plugin.json

@@ -3,7 +3,7 @@
   "name": "codebase-index",
   "displayName": "Codebase Index",
   "description": "Give Claude a precise local map of your codebase: find implementations, trace behavior, and predict change impact with file-line evidence.",
-  "version": "2.1.1",
+  "version": "2.1.2",
   "author": {
     "name": "codebase-index contributors"
   },

codebase_index-2.1.2/.codex/skills/codebase-index/.skill_version

@@ -0,0 +1 @@
+2.1.2

codebase_index-2.1.2/.opencode/skills/codebase-index/.skill_version

@@ -0,0 +1 @@
+2.1.2

{codebase_index-2.1.1 → codebase_index-2.1.2}/CHANGELOG.md

@@ -6,6 +6,22 @@ All notable changes to this project are documented here. The format is based on
 
 ## [Unreleased]
 
+## [2.1.2] - 2026-09-25
+
+### Fixed
+
+- The `PreToolUse` guard no longer blocks work it should let through:
+  - the first `codebase-index` command of a session turns the guard off even when it
+    is the command that builds the index; before, a repository without an index kept
+    the guard on for the whole session. Session state now lives in the system temp
+    directory (`CBX_HOOK_STATE` overrides it) instead of inside the index;
+  - heredoc bodies (`cat > script.ps1 <<'EOF'`, `python - <<EOF`) are file content, not
+    commands, and are no longer scanned for `grep`/`Select-String`;
+  - a retry is matched by its search term, not by the whole tool call, so a rewritten
+    command or a new description for the same search goes through;
+  - a quoted alternation such as `grep "a\|b"` is read whole, and a search on its own
+    line of a multi-line command is recognised.
+
 ## [2.1.1] - 2026-09-25
 
 ### Added
@@ -886,7 +902,8 @@ Pooled over 305 queries (Python, Java, TypeScript), v1.8.0 → 1.9.0:
 - Hooks example + `watch` mode for keeping the index fresh without blocking the edit loop (M8).
 - `doctor`, `stats`, `clean` diagnostics/maintenance commands.
 
-[Unreleased]: https://github.com/denfry/codebase-index/compare/v2.1.1...HEAD
+[Unreleased]: https://github.com/denfry/codebase-index/compare/v2.1.2...HEAD
+[2.1.2]: https://github.com/denfry/codebase-index/compare/v2.1.1...v2.1.2
 [2.1.1]: https://github.com/denfry/codebase-index/compare/v2.1.0...v2.1.1
 [2.1.0]: https://github.com/denfry/codebase-index/compare/v2.0.0...v2.1.0
 [2.0.0]: https://github.com/denfry/codebase-index/compare/v1.9.0...v2.0.0

{codebase_index-2.1.1 → codebase_index-2.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codebase-index
-Version: 2.1.1
+Version: 2.1.2
 Summary: A precise local codebase map for AI coding agents: find, trace, and predict with file-line evidence.
 Project-URL: Homepage, https://github.com/denfry/codebase-index
 Project-URL: Documentation, https://github.com/denfry/codebase-index/tree/main/docs

{codebase_index-2.1.1 → codebase_index-2.1.2}/docs/installer.md

@@ -110,7 +110,7 @@ pwsh ./install.ps1 -Target claude -InstallDir "D:\skills\codebase-index"
 **Pin to a branch or tag** (reproducibility and safety):
 
 ```sh
-sh install.sh --branch v2.1.1
+sh install.sh --branch v2.1.2
 ```
 
 ---

{codebase_index-2.1.1 → codebase_index-2.1.2}/requirements.lock

@@ -1,3 +1,3 @@
-codebase-index @ https://github.com/denfry/codebase-index/archive/refs/tags/v2.1.1.tar.gz
+codebase-index @ https://github.com/denfry/codebase-index/archive/refs/tags/v2.1.2.tar.gz
 tree-sitter==0.25.2
 tree-sitter-language-pack==1.8.1

{codebase_index-2.1.1 → codebase_index-2.1.2}/src/codebase_index/__init__.py

@@ -4,4 +4,4 @@ Public surface is intentionally small; the CLI (`codebase_index.cli:app`) is the
 See docs/ARCHITECTURE.md for the module map.
 """
 
-__version__ = "2.1.1"
+__version__ = "2.1.2"

{codebase_index-2.1.1 → codebase_index-2.1.2}/src/codebase_index/hooks.py

@@ -6,9 +6,11 @@ Two events, both no-ops outside a repository that has an index:
   is indexed, and which commands answer code questions.
 - ``guard`` (PreToolUse on Grep and Bash) intercepts a code search the first time
   in a session, before any ``codebase-index`` command has run, and names the
-  index command to use instead. It is a nudge, not a wall: repeating the same
-  call lets it through, the first ``codebase-index`` command in the session turns
-  the guard off, and searches over docs, logs or config are never intercepted.
+  index command to use instead. It is a nudge, not a wall: searching for the
+  same term again lets it through, the first ``codebase-index`` command in the
+  session turns the guard off (even one that has yet to build the index), and
+  searches over docs, logs or config, or inside heredoc bodies, are never
+  intercepted.
 
 Measured on a 5.9k-file monorepo, an agent spent 19% fewer tokens with the index
 than with grep (tests/eval/results/2026-09-24-agent-pilot.md), but agents reach
@@ -17,6 +19,8 @@ only the standard library and does no index work beyond one small SQLite read.
 
 Entry point: ``codebase-index-hook <event>`` reads the hook JSON on stdin and
 prints the hook JSON reply (or nothing). ``CBX_GUARD=0`` disables the guard.
+Per-session state lives in the system temp directory (``CBX_HOOK_STATE``
+overrides it), so it exists before the repository has an index.
 """
 
 from __future__ import annotations
@@ -27,21 +31,31 @@ import os
 import re
 import sqlite3
 import sys
+import tempfile
 import time
 from pathlib import Path
 from typing import Any, Optional
 
 INDEX_REL = Path(".claude") / "cache" / "codebase-index" / "index.sqlite"
-_STATE_DIR = "hook-sessions"
+_STATE_DIR = "codebase-index-hooks"
 _STATE_TTL_S = 2 * 24 * 3600
 
 # A shell segment that searches file contents: the first command of the segment
-# (after `&&`, `||`, `;` or the start), so `ps aux | grep x` is not a repo search.
+# (after `&&`, `||`, `;`, a newline or the start), so `ps aux | grep x` is not a
+# repo search. Quoted arguments may hold `|`, as in grep "a\|b".
 _SEARCH_CMD = re.compile(
-    r"(?:^|&&|\|\||;)\s*(?:cd\s+\S+\s*&&\s*)?"
-    r"(?P<cmd>git\s+grep|grep|egrep|fgrep|rg|ag|ack|findstr|Select-String)\b(?P<args>[^|;&]*)",
+    r"(?:^|&&|\|\||;|\n)[ \t]*(?:cd\s+\S+\s*&&\s*)?"
+    r"(?P<cmd>git\s+grep|grep|egrep|fgrep|rg|ag|ack|findstr|Select-String)\b"
+    r"""(?P<args>(?:"[^"\n]*"|'[^'\n]*'|[^|;&\n"'])*)""",
     re.IGNORECASE,
 )
+# A heredoc body is file content (a script being written, stdin for python), not
+# commands; an unterminated one runs to the end of the command.
+_HEREDOC = re.compile(
+    r"""(?<!<)<<-?(?!<)[ \t]*(['"]?)([A-Za-z_][\w-]*)\1(?P<rest>[^\n]*)\n"""
+    r".*?(?:^[ \t]*\2[ \t]*$|\Z)",
+    re.DOTALL | re.MULTILINE,
+)
 _INDEX_CMD = re.compile(r"(?:^|[\s;&|/\\\"'])(?:codebase-index|cbx)(?:\.exe|\.ps1)?\s+\w")
 # Searches over prose, logs and config are what grep is for; leave them alone.
 _NON_CODE = re.compile(
@@ -132,15 +146,12 @@ def guard(payload: dict) -> Optional[dict]:
     tool_input = payload.get("tool_input") or {}
     if tool not in ("Grep", "Bash"):
         return None
-    root = find_index_root(_cwd(payload))
-    if root is None:
-        return None
-    session = str(payload.get("session_id") or "default")
-    state = _State(root, session)
+    state = _State(str(payload.get("session_id") or "default"))
 
     if tool == "Bash":
-        command = str(tool_input.get("command") or "")
+        command = _HEREDOC.sub(lambda m: "<<" + m.group("rest"), str(tool_input.get("command") or ""))
         if _INDEX_CMD.search(command):
+            # Checked before the index exists: the first search is what builds it.
             state.mark_used()
             return None
         pattern = _shell_search_pattern(command)
@@ -151,11 +162,11 @@ def guard(payload: dict) -> Optional[dict]:
             return None
         pattern = str(tool_input.get("pattern") or "")
 
-    if state.used:
+    if state.used or find_index_root(_cwd(payload)) is None:
         return None
-    fingerprint = hashlib.sha1(
-        (tool + json.dumps(tool_input, sort_keys=True, default=str)).encode("utf-8")
-    ).hexdigest()
+    # Keyed by the search term, not the whole call: a retry rarely repeats the
+    # command byte for byte (a new description, another pipe, a regenerated script).
+    fingerprint = hashlib.sha1(_term(pattern).lower().encode("utf-8")).hexdigest()
     if state.was_denied(fingerprint):
         return None  # the retry after a nudge: the agent decided grep is right
     state.deny(fingerprint)
@@ -190,26 +201,31 @@ def _grep_targets_non_code(tool_input: dict) -> bool:
     return bool(_NON_CODE.search(target))
 
 
+def _term(pattern: str) -> str:
+    """A regex search pattern as plain query words."""
+    return " ".join(re.sub(r"""[\\^$()\[\]{}|*+?"']""", " ", pattern).split())[:80]
+
+
 def _reason(pattern: str) -> str:
-    term = re.sub(r"[\\^$()\[\]{}|*+?]", " ", pattern).strip() or "<what you are looking for>"
-    term = " ".join(term.split())[:80]
+    term = _term(pattern) or "<what you are looking for>"
     return (
         "codebase-index: this repository is indexed, and the index answers code "
         "searches with ranked, numbered lines in fewer tokens than grep. Run it first:\n"
         f'  codebase-index search "{term}" --compact     # where / how\n'
         '  codebase-index refs "Owner.member" --compact  # every call site, with callers\n'
         '  codebase-index symbol "Name"                 # a definition\n'
-        "If the index does not answer, repeat this exact call and it will run. Once any "
-        "codebase-index command has run in this session, searches are not intercepted."
+        "If the index does not answer, search for the same term again and it will run. "
+        "Once any codebase-index command has run in this session, searches are not "
+        "intercepted."
     )
 
 
 class _State:
     """Per-session guard state: whether the index was used, which calls were nudged."""
 
-    def __init__(self, root: Path, session: str) -> None:
+    def __init__(self, session: str) -> None:
         safe = re.sub(r"[^A-Za-z0-9_.-]", "_", session)[:80] or "default"
-        self.dir = root / INDEX_REL.parent / _STATE_DIR
+        self.dir = Path(os.environ.get("CBX_HOOK_STATE") or Path(tempfile.gettempdir()) / _STATE_DIR)
         self.path = self.dir / f"{safe}.json"
         self.data: dict[str, Any] = {"used": False, "denied": []}
         try:

{codebase_index-2.1.1 → codebase_index-2.1.2}/tests/test_hooks_guard.py

@@ -11,6 +11,11 @@ import pytest
 from codebase_index import hooks, scaffold
 
 
+@pytest.fixture(autouse=True)
+def _state_dir(tmp_path: Path, monkeypatch):
+    monkeypatch.setenv("CBX_HOOK_STATE", str(tmp_path / "hook-state"))
+
+
 @pytest.fixture
 def repo(tmp_path: Path) -> Path:
     db = tmp_path / "repo" / hooks.INDEX_REL
@@ -68,6 +73,38 @@ def test_non_code_and_non_search_shell_commands_pass(repo: Path, command: str):
     assert _call(repo, "Bash", command=command) is None
 
 
+def test_the_retry_is_matched_by_search_term_not_by_the_whole_call(repo: Path):
+    assert _denied(_call(repo, "Bash", command="grep -rn TownService src",
+                         description="Find the service"))
+    assert _call(repo, "Bash", command="cd src && grep -rn TownService . | head",
+                 description="Find it again") is None
+
+
+def test_an_index_command_that_builds_the_index_turns_the_guard_off(tmp_path: Path, repo: Path):
+    # The first `codebase-index search` in a fresh repository builds the index, so the
+    # guard sees that call before any index exists.
+    fresh = tmp_path / "fresh"
+    (fresh / "src").mkdir(parents=True)
+    assert _call(fresh, "Bash", command='codebase-index search "x" --compact') is None
+    assert _call(repo, "Grep", pattern="anything") is None
+
+
+@pytest.mark.parametrize("command", [
+    # A heredoc body is file content, not commands.
+    "cat > /tmp/close.ps1 <<'EOF'\n$log = Get-Content x; Select-String \"BUILD FAILED\" $log\nEOF\n"
+    "echo done",
+    "python - <<EOF\nimport re; grep = 1\nEOF",
+])
+def test_heredoc_bodies_are_not_searched(repo: Path, command: str):
+    assert _call(repo, "Bash", command=command) is None
+
+
+def test_a_search_on_its_own_line_is_guarded_and_quoted_alternation_kept(repo: Path):
+    reply = _call(repo, "Bash", command='ls src\ngrep -rn "Lifecycle\\|onInit" src --include=*.java')
+    assert _denied(reply)
+    assert 'search "Lifecycle onInit"' in reply["hookSpecificOutput"]["permissionDecisionReason"]
+
+
 @pytest.mark.parametrize("command", [
     "grep -rn treasurerDeparted --include=*.java .",
     "cd /c/Projects/x && rg -n 'TownService.refresh' realism-polity",

codebase_index-2.1.1/.claude/skills/codebase-index/.skill_version

@@ -1 +0,0 @@
-2.1.1

codebase_index-2.1.1/.codex/skills/codebase-index/.skill_version

@@ -1 +0,0 @@
-2.1.1

codebase_index-2.1.1/.opencode/skills/codebase-index/.skill_version

@@ -1 +0,0 @@
-2.1.1