claude-dev-env 8.100.0 → 8.100.1

package/hooks/blocking/test_verify_before_commit_gate.py

@@ -43,13 +43,32 @@ TOOL_RESULT_ENTRY = {
     },
 }
 
+REFUSED_TOOL_RESULT_ENTRY = {
+    "type": "user",
+    "message": {
+        "content": [
+            {
+                "type": "tool_result",
+                "tool_use_id": "toolu_1",
+                "content": "<tool_use_error>Skill verify cannot be used with Skill tool</tool_use_error>",
+                "is_error": True,
+            }
+        ]
+    },
+}
+
 
 def _skill_entry(skill_name: str) -> dict[str, object]:
     return {
         "type": "assistant",
         "message": {
             "content": [
-                {"type": "tool_use", "name": "Skill", "input": {"skill": skill_name}}
+                {
+                    "type": "tool_use",
+                    "id": "toolu_1",
+                    "name": "Skill",
+                    "input": {"skill": skill_name},
+                }
             ]
         },
     }
@@ -197,6 +216,33 @@ def test_docs_only_commit_passes(repository: Path, tmp_path: Path) -> None:
     )
 
 
+def test_data_files_under_docs_folder_pass(repository: Path, tmp_path: Path) -> None:
+    (repository / "docs" / "records").mkdir(parents=True)
+    (repository / "docs" / "records" / "proof.json").write_text("{}\n", encoding="utf-8")
+    (repository / "docs" / "ruleset.yaml").write_text("rules: []\n", encoding="utf-8")
+    _git(repository, "add", "-A")
+    assert not _is_denied(
+        _payload("git commit -m records", repository, _unverified(tmp_path))
+    )
+
+
+def test_code_under_docs_folder_is_denied(repository: Path, tmp_path: Path) -> None:
+    (repository / "docs").mkdir()
+    (repository / "docs" / "conf.py").write_text("VALUE = 2\n", encoding="utf-8")
+    _git(repository, "add", "-A")
+    assert _is_denied(
+        _payload("git commit -m conf", repository, _unverified(tmp_path))
+    )
+
+
+def test_json_outside_docs_folder_is_denied(repository: Path, tmp_path: Path) -> None:
+    (repository / "config.json").write_text("{}\n", encoding="utf-8")
+    _git(repository, "add", "-A")
+    assert _is_denied(
+        _payload("git commit -m config", repository, _unverified(tmp_path))
+    )
+
+
 def test_tests_only_commit_passes(repository: Path, tmp_path: Path) -> None:
     (repository / "test_app.py").write_text(
         "def test_value() -> None:\n    pass\n", encoding="utf-8"
@@ -219,6 +265,15 @@ def test_docs_plus_code_commit_is_denied(repository: Path, tmp_path: Path) -> No
     )
 
 
+def test_refused_verify_skill_call_is_denied(repository: Path, tmp_path: Path) -> None:
+    (repository / "app.py").write_text("VALUE = 2\n", encoding="utf-8")
+    _git(repository, "add", "app.py")
+    transcript_path = _transcript(
+        tmp_path, PROMPT_ENTRY, _skill_entry("verify"), REFUSED_TOOL_RESULT_ENTRY
+    )
+    assert _is_denied(_payload("git commit -m change", repository, transcript_path))
+
+
 def test_verify_skill_call_after_the_last_prompt_passes(
     repository: Path, tmp_path: Path
 ) -> None:

package/hooks/blocking/verify_before_commit_gate.py

@@ -33,6 +33,8 @@ from hooks_constants.verify_before_commit_gate_constants import (
     ALL_ADD_EVERYTHING_OPTIONS,
     ALL_COMMIT_ALL_OPTIONS,
     ALL_COMMIT_VALUED_SHORT_FLAGS,
+    ALL_DOCS_DATA_EXTENSIONS,
+    ALL_DOCS_DIRECTORY_NAMES,
     ALL_DOCS_EXTENSIONS,
     ALL_GIT_EXECUTABLES,
     ALL_SLASH_COMMAND_MARKERS,
@@ -176,7 +178,12 @@ def _is_docs_or_test_path(repository_path: str) -> bool:
     path = PurePosixPath(repository_path)
     if path.suffix.lower() in ALL_DOCS_EXTENSIONS or path.name in ALL_TEST_FILE_NAMES:
         return True
-    if any(each_part in ALL_TEST_DIRECTORY_NAMES for each_part in path.parts[:-1]):
+    all_directory_parts = path.parts[:-1]
+    if path.suffix.lower() in ALL_DOCS_DATA_EXTENSIONS and any(
+        each_part in ALL_DOCS_DIRECTORY_NAMES for each_part in all_directory_parts
+    ):
+        return True
+    if any(each_part in ALL_TEST_DIRECTORY_NAMES for each_part in all_directory_parts):
         return True
     return (
         path.name.startswith(ALL_TEST_FILE_PREFIXES)
@@ -245,6 +252,9 @@ def decision_for(all_payload_fields: dict[str, object]) -> dict[str, object] | N
         git commit, code staged, no verify since the last prompt      -> DENY_REASON
         git add app.py; git commit, empty index, no verify            -> DENY_REASON
         git commit, only docs or test files                           -> None
+        git commit, only data files (.json, .jsonl, .yaml, .yml,
+            .csv) under a docs folder                                 -> None
+        git commit, a .py file under a docs folder                    -> DENY_REASON
         git commit after a verify Skill call or a /verify prompt      -> None
         git commit in a repository with no verify skill directory     -> None
         transcript missing or unreadable, or git fails                -> None

package/hooks/features/conduct-gates.md

@@ -18,7 +18,7 @@ This family stops a chat reply, edited message, or tool call when its payload br
 - `blocking/verify_before_acting.py` blocks a completed mutating call when the agent's reasoning contains an unchecked claim.
 - `blocking/pr_lifecycle_skill_gate.py` denies a commit, push, pull request action, or merge until the `pr-lifecycle` skill appears in the transcript since the last compaction.
 - `blocking/resolve_thread_reply_gate.py` denies a review thread resolve until the agent's own transcript shows a reply on that same thread that posted after the last user prompt and that no successful resolve of the thread has spent. A resolve is a `resolve_review_thread` call, which names the thread by `threadID`; a `pull_request_review_write` call with method `resolve_thread`, which names it by `threadId`; a shell command with a `gh api graphql` `resolveReviewThread` mutation, which names it by a literal `PRRT_` id; or a shell call to `pulls/<n>/ccr/comments/<id>/resolve`, which names it by a comment id. A graphql resolve that names no `PRRT_` id is denied. A reply is an `add_reply_to_pull_request_comment` call, or a `gh api` call to a comment's `replies` endpoint or to `addPullRequestReviewThreadReply`, which names its thread by `PRRT_` id. A reply and a resolve by comment id match when they name the same comment, or when a successful tool result in the transcript, such as `pull_request_read` `get_review_comments`, lists the comment under the thread by its `discussion_r` anchor or `databaseId`. A reply whose result is an error, such as a refusal from `blocking/reply_length_gate.py`, does not count. With no readable transcript, the resolve passes.
-- `blocking/verify_before_commit_gate.py` denies a `git commit` in a repository with a `.claude/skills/verify` or `.agents/skills/verify` directory until the `verify` skill loads after the last user prompt, through a Skill call or the `/verify` command in that prompt. It reads the committed paths from the staged diff, the tracked changes for `-a`, and the paths of each `git add` earlier in the same command, and passes a commit whose paths are all docs, images, or test files.
+- `blocking/verify_before_commit_gate.py` denies a `git commit` in a repository with a `.claude/skills/verify` or `.agents/skills/verify` directory until the `verify` skill loads after the last user prompt, through a Skill call whose result comes back without an error, or the `/verify` command in that prompt. It reads the committed paths from the staged diff, the tracked changes for `-a`, and the paths of each `git add` earlier in the same command, and passes a commit whose paths are all docs, images, test files, or `.json`, `.jsonl`, `.yaml`, `.yml`, or `.csv` files under a `docs` folder.
 - `blocking/build_eval_before_eval_write_gate.py` denies a write to an eval file until a Skill call named exactly `claude-api` with args `build-eval`, or the `/claude-api build-eval` command, appears in the transcript.
 - `blocking/merge_review_gate.py` denies an agent's merge until a review pass on the merged head reported zero findings, or ruled out every finding and linked the posted rulings. It reads the receipt `scripts/merge_review_receipt.py` writes, requires the merge, auto-merge or merge-queue entry to be pinned to that head, and skips repositories listed in `CLAUDE_MERGE_REVIEW_EXEMPT_REPOSITORIES`. Its deny reason names the review command `CLAUDE_MERGE_REVIEW_COMMAND` sets. The reason tells the agent to run that command for the first three passes and `/code-review low --fix <number>` for each later pass; the gate itself counts no passes. It also denies a pull request open, through `gh pr create`, `pull_request.py create` or a `create_pull_request` tool, until a pass recorded as `merge_review_receipt.py <owner/name> open <head> 0` covers the head the pull request opens with. A merge pinned to that same head passes on the open receipt.
 - `blocking/merge_over_review_offer_gate.py` denies a decision card or question that offers a merge while it names open review findings, a red check, unaddressed review comments, or an override. It reads the whole card as one text: every question, header, context, reason, and option, whether the option is a string or a label with a consequence or description.
@@ -68,7 +68,7 @@ Preconditions:
 - **Step note.** Input is a tool call after a transcript message without a status line while the flag is on. Run `python -m pytest packages/claude-dev-env/hooks/blocking/test_step_note_gate.py -q`. The adjacent test observes a block until a status line appears.
 - **Checked claim.** Input is a Write after hedged reasoning in the transcript. Run `python -m pytest packages/claude-dev-env/hooks/blocking/test_verify_before_acting.py -q`. The adjacent test observes a block and a logged blocked outcome.
 - **Lifecycle skill.** Input is a `git commit` or `gh pr create` command with no `pr-lifecycle` invocation in the transcript. Run `python -m pytest packages/claude-dev-env/hooks/blocking/test_pr_lifecycle_skill_gate.py -q`. The adjacent test observes a denial, then a silent allow once the skill or its slash command appears.
-- **Verify before commit.** Input is `git add app.py; git commit -m change` in a temporary repository with `.agents/skills/verify`, and a transcript with no `verify` load after its last prompt. Run `python -m pytest packages/claude-dev-env/hooks/blocking/test_verify_before_commit_gate.py -q`. The adjacent test observes a denial, then a silent allow after a `verify` Skill call or a `/verify` prompt. A load before the last prompt still denies. A docs-only commit, a tests-only commit, and a repository with no verify skill pass.
+- **Verify before commit.** Input is `git add app.py; git commit -m change` in a temporary repository with `.agents/skills/verify`, and a transcript with no `verify` load after its last prompt. Run `python -m pytest packages/claude-dev-env/hooks/blocking/test_verify_before_commit_gate.py -q`. The adjacent test observes a denial, then a silent allow after a `verify` Skill call or a `/verify` prompt. A load before the last prompt still denies. A docs-only commit, a commit of data files under a `docs` folder, a tests-only commit, and a repository with no verify skill pass. A `.py` file under `docs` and a `.json` file outside `docs` still deny.
 - **Eval write.** Input is a Write to `evals/cases.json` with no build-eval call in the transcript. Run `python -m pytest packages/claude-dev-env/hooks/blocking/test_build_eval_before_eval_write_gate.py -q`. The adjacent test observes a denial, then an allow once the call appears; a write outside an eval path passes.
 - **Clean review.** Input is a merge with no receipt, a receipt with findings, a merge pinned to a newer head, an unpinned merge, or auto-merge with no head. Run `python -m pytest packages/claude-dev-env/hooks/blocking/test_merge_review_gate.py -q`. The adjacent test observes a denial for each; a merge or auto-merge pinned to the clean head and a merge in an exempt repository pass. An open with no clean open receipt at its head is denied; an open reviewed clean at its head, and a merge pinned to that head, pass.
 - **Merge offer.** Input is a decision card or AskUserQuestion that offers "Merge now", the merge queue, or a yes to "Merge it now?" while any part of the card names open findings or unaddressed review comments. Run `python -m pytest packages/claude-dev-env/hooks/blocking/test_merge_over_review_offer_gate.py -q`. The adjacent test observes a denial for each, including an offer and findings in separate questions; a card with no merge offer, a card about the merge review check itself, a card about shipping a docs change, and a merge offer after a clean review pass.

package/hooks/hooks_constants/verify_before_commit_gate_constants.py

@@ -3,7 +3,7 @@
 DENY_REASON = (
     "This repository has a verify skill, and this commit changes code. Invoke the verify "
     "skill with the Skill tool, run the check it routes you to, then run the same command again. "
-    "A commit that changes only docs or tests passes without it."
+    "A commit that changes only docs, data files under a docs folder, or tests passes without it."
 )
 SKILL_NAME = "verify"
 ALL_SLASH_COMMAND_MARKERS = ("<command-name>/verify</command-name>",)
@@ -47,6 +47,8 @@ ALL_DOCS_EXTENSIONS = frozenset(
         ".ico",
     }
 )
+ALL_DOCS_DIRECTORY_NAMES = frozenset({"docs"})
+ALL_DOCS_DATA_EXTENSIONS = frozenset({".json", ".jsonl", ".yaml", ".yml", ".csv"})
 ALL_TEST_DIRECTORY_NAMES = frozenset({"tests", "test", "__tests__"})
 ALL_TEST_FILE_PREFIXES = ("test_",)
 ALL_TEST_FILE_SUFFIXES = ("_test.py",)

package/hooks/test_transcript_skill_scan.py

@@ -18,9 +18,14 @@ from transcript_skill_scan import (
 )
 
 
-def _skill_entry(name: str) -> str:
+def _skill_entry(name: str, tool_use_id: str = "toolu_skill") -> str:
     return json.dumps(
-        {"type": "assistant", "message": {"content": [{"type": "tool_use", "name": "Skill", "input": {"skill": name}}]}}
+        {
+            "type": "assistant",
+            "message": {
+                "content": [{"type": "tool_use", "id": tool_use_id, "name": "Skill", "input": {"skill": name}}]
+            },
+        }
     )
 
 
@@ -144,12 +149,40 @@ def _prompt_entry(content: object) -> str:
 
 
 def test_skill_since_last_prompt_survives_tool_results_and_skill_bodies() -> None:
-    tool_result = _prompt_entry([{"type": "tool_result", "content": "Launching skill: verify"}])
+    tool_result = _prompt_entry(
+        [{"type": "tool_result", "tool_use_id": "toolu_skill", "content": "Launching skill: verify"}]
+    )
     skill_body = json.dumps({"type": "user", "isMeta": True, "message": {"content": "Base directory for this skill"}})
     lines = [_prompt_entry("Commit it."), _skill_entry("verify"), tool_result, skill_body]
     assert is_skill_loaded_since_last_prompt(lines, ("verify",), ())
 
 
+def test_refused_skill_load_leaves_the_skill_unloaded() -> None:
+    refused_result = _prompt_entry(
+        [
+            {
+                "type": "tool_result",
+                "tool_use_id": "toolu_skill",
+                "content": "<tool_use_error>Skill verify cannot be used with Skill tool</tool_use_error>",
+                "is_error": True,
+            }
+        ]
+    )
+    lines = [_prompt_entry("Commit it."), _skill_entry("verify"), refused_result]
+    assert not is_skill_loaded_since_last_prompt(lines, ("verify",), ())
+
+
+def test_skill_call_with_no_result_yet_leaves_the_skill_unloaded() -> None:
+    lines = [_prompt_entry("Commit it."), _skill_entry("verify")]
+    assert not is_skill_loaded_since_last_prompt(lines, ("verify",), ())
+
+
+def test_result_for_another_call_leaves_the_skill_unloaded() -> None:
+    other_result = _prompt_entry([{"type": "tool_result", "tool_use_id": "toolu_other", "content": "done"}])
+    lines = [_prompt_entry("Commit it."), _skill_entry("verify"), other_result]
+    assert not is_skill_loaded_since_last_prompt(lines, ("verify",), ())
+
+
 def test_next_prompt_drops_a_skill_loaded_in_the_last_turn() -> None:
     lines = [_prompt_entry("First."), _skill_entry("verify"), _prompt_entry([{"type": "text", "text": "Second."}])]
     assert not is_skill_loaded_since_last_prompt(lines, ("verify",), ())

package/hooks/transcript_skill_scan.py

@@ -44,6 +44,17 @@ def _named_skill_inputs(
     all_entry_fields: dict[str, object],
     skill_names: Collection[str],
     accepts_plugin_prefix: bool = True,
+) -> Iterator[dict[str, object]]:
+    for each_block in _named_skill_blocks(all_entry_fields, skill_names, accepts_plugin_prefix):
+        skill_input = each_block["input"]
+        if isinstance(skill_input, dict):
+            yield skill_input
+
+
+def _named_skill_blocks(
+    all_entry_fields: dict[str, object],
+    skill_names: Collection[str],
+    accepts_plugin_prefix: bool = True,
 ) -> Iterator[dict[str, object]]:
     message = all_entry_fields.get("message")
     if all_entry_fields.get("type") != ASSISTANT_ENTRY_TYPE or not isinstance(message, dict):
@@ -61,7 +72,7 @@ def _named_skill_inputs(
             invoked_name == name or (accepts_plugin_prefix and invoked_name.endswith(":" + name))
             for name in skill_names
         ):
-            yield each_block["input"]
+            yield each_block
 
 
 def _invokes_skill(
@@ -183,11 +194,15 @@ def is_skill_loaded_since_last_prompt(
     ::
 
         prompt, Skill verify, tool result, skill body   -> True
+        prompt, Skill verify, tool result with is_error -> False
+        prompt, Skill verify, no tool result yet        -> False
         prompt, Skill verify, prompt                    -> False
         prompt carrying the /verify command marker      -> True
         prompt with the marker, compact boundary        -> False
 
-    A tool result and a meta entry, such as the skill body, leave the turn open.
+    A Skill call counts once its tool result comes back without an error, so a
+    refused load leaves the skill unloaded. A tool result and a meta entry, such
+    as the skill body, leave the turn open.
 
     Args:
         all_transcript_lines: JSON transcript entries, one per line.
@@ -195,13 +210,23 @@ def is_skill_loaded_since_last_prompt(
         slash_command_markers: Accepted user command tags.
     """
     loaded = False
+    all_pending_ids: set[object] = set()
     for each_line in all_transcript_lines:
         all_entry_fields = _parsed_entry(each_line)
         if all_entry_fields is None:
             continue
         if all_entry_fields.get("subtype") == COMPACT_BOUNDARY_SUBTYPE or _is_user_prompt(all_entry_fields):
             loaded = False
-        loaded = loaded or _invokes_skill(all_entry_fields, skill_names, slash_command_markers)
+            all_pending_ids = set()
+        command_text = _user_command_text(all_entry_fields)
+        if command_text is not None:
+            loaded = loaded or any(marker in command_text for marker in slash_command_markers)
+        all_pending_ids.update(
+            each_block["id"]
+            for each_block in _named_skill_blocks(all_entry_fields, skill_names)
+            if each_block.get("id") is not None
+        )
+        loaded = loaded or bool(all_pending_ids & _succeeded_result_ids(all_entry_fields))
     return loaded
 
 

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "claude-dev-env",
-    "version": "8.100.0",
+    "version": "8.100.1",
     "description": "Claude Code development standards — rules, hooks, agents, commands, and skills",
     "type": "module",
     "bin": {