cecli-dev 1.8.1rc1__tar.gz → 1.8.1rc2__tar.gz

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cecli-dev
-Version: 1.8.1rc1
+Version: 1.8.1rc2
 Summary: we also can't pronounce cecli
 Project-URL: Homepage, https://github.com/dwash96/cecli
 Project-URL: Documentation, https://github.com/dwash96/cecli

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/cecli/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.8.1rc1'
-__version_tuple__ = version_tuple = (1, 8, 1, 'rc1')
+__version__ = version = '1.8.1rc2'
+__version_tuple__ = version_tuple = (1, 8, 1, 'rc2')
 
-__commit_id__ = commit_id = 'g15af71fcd'
+__commit_id__ = commit_id = 'g9c5114f1f'

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/cecli/coders/base_coder.py

@@ -2253,6 +2253,31 @@ class Coder(metaclass=UsageMeta):
                 )
                 return
 
+        # Compaction can only shrink the conversation buckets counted above.
+        # If the fixed context (files added to the chat, repo map, system
+        # prompt, rules, tool schemas) already meets the limit on its own,
+        # summarizing cannot bring the context under the limit. Bail out with
+        # guidance instead of churning every turn, which is what a large file
+        # added to the chat otherwise causes.
+        non_compactable_tokens = all_tokens - compactable_tokens
+
+        if (
+            self.context_compaction_max_tokens
+            and non_compactable_tokens >= self.context_compaction_max_tokens
+        ):
+            if getattr(self, "_compaction_noop_warning_time", 0) + 300 < time.time():
+                self._compaction_noop_warning_time = time.time()
+                self.io.tool_warning(
+                    "Compaction cannot reduce the context: the fixed context (files in the"
+                    f" chat, repo map, prompts) is already ~{non_compactable_tokens:,}"
+                    f" tokens, at or above the ~{self.context_compaction_max_tokens:,} token"
+                    " compaction limit."
+                )
+                self.io.tool_output("- Use /drop to remove unneeded files from the chat")
+                self.io.tool_output("- Use /clear to clear the chat history")
+
+            return
+
         message_tokens = done_tokens + cur_tokens
         file_tokens = diff_tokens + file_context_tokens
         combined_tokens = done_tokens + cur_tokens + diff_tokens + file_context_tokens

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/cecli/helpers/llms/litellm_compat.py

@@ -19,6 +19,7 @@ from __future__ import annotations
 
 import asyncio
 import json
+import math
 import os
 import re
 import warnings
@@ -1096,15 +1097,11 @@ class _LiteLLMFacade:
         """Estimate tokens for ``text`` without loading a BPE tokenizer.
 
         tiktoken's cl100k_base/o200k_base encodings cost ~30-95MB of RSS the
-        first time they are loaded, so we approximate instead: roughly one token
-        per four ASCII characters (the classic rule of thumb), plus one token per
-        non-ASCII character (CJK/emoji tokenize at about one token per char), and
-        never fewer tokens than there are whitespace-delimited word/punctuation
-        runs (dense code with short tokens is otherwise undercounted).
-
-        The estimate lands within ~±30% of cl100k_base on code and prose, which
-        is plenty for context-window checks, cost display, and file-size
-        warnings.
+        first time they are loaded, so we approximate instead with a regex
+        pre-tokenizer patterned on OpenAI's o200k_base (see
+        ``_estimate_token_count``). The estimate lands within ~±30% of
+        cl100k_base on code and prose, which is plenty for context-window
+        checks, cost display, and file-size warnings.
 
         Returns a ``range`` whose length is the estimated token count; the
         individual values are dummy ids (callers only use ``len()``/iteration).
@@ -1244,23 +1241,41 @@ __all__ = ["litellm"]
 # Token-count estimator (tiktoken-free)
 # ---------------------------------------------------------------------------
 
-_WORD_OR_PUNCT_RE = re.compile(r"\w+|[^\w\s]")
-_NON_ASCII_RE = re.compile(r"[^\x00-\x7f]")
+# Standard ``re`` approximation of OpenAI's o200k_base pre-tokenizer pattern.
+# The letter class is Unicode-aware (``[^\W\d_]``) so non-Latin scripts are
+# counted rather than dropped.
+_O200K_APPROX_PATTERN = re.compile(
+    r"""'(?:[sdmt]|ll|ve|re)| ?[^\W\d_]+| ?\d{1,3}| ?[^\s\w]+|\s+""", re.UNICODE
+)
+
+#: Bytes charged per token for chunks too long to be a single token, calibrated
+#: against o200k_base (~5 bytes/token on code, prose and JSON).
+_CHARS_PER_TOKEN = 5.0
 
 
 def _estimate_token_count(text: str) -> int:
     """Estimate the number of tokens in ``text`` without a BPE tokenizer.
 
-    Uses the classic ~4 chars/token rule of thumb, lifted for non-ASCII text
-    (CJK/emoji tokenize at roughly one token per character) and floored by the
-    number of word/punctuation runs so dense code with many short tokens isn't
-    undercounted.
+    Splits ``text`` with a regex approximation of OpenAI's o200k_base
+    pre-tokenizer, then accumulates a fractional token weight per chunk:
+    whitespace runs collapse to a single token, numbers and short chunks
+    (<= 4 bytes) count as one, and longer chunks are charged at roughly
+    ``_CHARS_PER_TOKEN`` bytes per token. The total is rounded up and floored
+    at one token for any non-empty input.
     """
 
     if not text:
         return 0
 
-    units = len(_WORD_OR_PUNCT_RE.findall(text))
-    non_ascii = len(_NON_ASCII_RE.findall(text))
+    tokens = 0.0
+    for chunk in _O200K_APPROX_PATTERN.findall(text):
+        byte_len = len(chunk.encode("utf-8"))
+
+        if chunk.isspace():
+            tokens += 1.0
+        elif chunk.strip().isdigit() or byte_len <= 4:
+            tokens += 1.0
+        else:
+            tokens += max(1.0, byte_len / _CHARS_PER_TOKEN)
 
-    return max(1, len(text) // 4, units) + non_ascii
+    return max(1, math.ceil(tokens - 1e-9))

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/cecli/helpers/workflows/config.py

@@ -11,8 +11,9 @@ Resolution order, highest precedence first:
    lower-precedence sources still load.
 2. ``<git_root or cwd>/.cecli.workflows.yml`` / ``.cecli.workflows.yaml``.
 3. ``<git_root or cwd>/.cecli/workflows/*.yml`` / ``*.yaml`` — globbed in stable sorted order.
-4. ``merged_config["workflows"]`` — the deep-merged ``.cecli.conf.yml`` value.
-5. ``~/.cecli/workflows/*.yml`` / ``*.yaml`` — globbed in stable sorted order.
+4. ``<git_root or cwd>/.agents/workflows/*.yml`` / ``*.yaml`` — globbed in stable sorted order.
+5. ``merged_config["workflows"]`` — the deep-merged ``.cecli.conf.yml`` value.
+6. ``~/.cecli/workflows/*.yml`` / ``*.yaml`` — globbed in stable sorted order.
 
 Each source may hold a single workflow mapping or a ``workflows: [...]`` list.
 On a duplicate ``name`` the higher-precedence definition wins and the loser is
@@ -42,6 +43,8 @@ WORKFLOW_FILENAMES = (".cecli.workflows.yml", ".cecli.workflows.yaml")
 
 WORKFLOW_DIR = Path(".cecli") / "workflows"
 
+AGENTS_WORKFLOW_DIR = Path(".agents") / "workflows"
+
 # Values argparse/deep-merge normalization leaves behind when nothing was set.
 _ABSENT_STRINGS = ("", "{}", "[]", "null", "~")
 
@@ -95,6 +98,7 @@ def resolve_workflow_configs(
 
     ordered += _entries_from_local(git_root)
     ordered += _entries_from_local_dir(git_root)
+    ordered += _entries_from_local_agents_dir(git_root)
 
     if isinstance(merged_config, dict):
         ordered += _entries(merged_config.get("workflows"), None, _CONFIG_LABEL)
@@ -298,6 +302,13 @@ def _entries_from_local_dir(git_root: Optional[Any]) -> List[_Discovered]:
     return _entries_from_dir(base / WORKFLOW_DIR)
 
 
+def _entries_from_local_agents_dir(git_root: Optional[Any]) -> List[_Discovered]:
+    """Resolve a repo-local ``.agents/workflows/*.yml|*.yaml`` directory."""
+    base = Path(git_root).expanduser() if git_root else Path.cwd()
+
+    return _entries_from_dir(base / AGENTS_WORKFLOW_DIR)
+
+
 def _entries_from_home() -> List[_Discovered]:
     """Resolve ``~/.cecli/workflows/*.yml|*.yaml`` in stable sorted order."""
 

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/cecli/website/docs/config/workflows.md

@@ -17,13 +17,14 @@ Preferred: one YAML file per workflow in `~/.cecli/workflows/`:
     morning-triage.yml
 ```
 
-Also supported: a repo-local `.cecli/workflows/` directory (one YAML file per workflow), inline under a `workflows:` key in `.cecli.conf.yml`, or a repo-local `.cecli.workflows.yml`. Each source may hold a single workflow mapping, a `workflows: [...]` list, or a plain list of mappings. Resolution precedence (highest first):
+Also supported: a repo-local `.cecli/workflows/` or `.agents/workflows/` directory (one YAML file per workflow), inline under a `workflows:` key in `.cecli.conf.yml`, or a repo-local `.cecli.workflows.yml`. Each source may hold a single workflow mapping, a `workflows: [...]` list, or a plain list of mappings. Resolution precedence (highest first):
 
 1. **CLI argument** — an inline JSON/YAML string or a file path passed to `--workflows`.
 2. **Local file** — `.cecli.workflows.yml` / `.cecli.workflows.yaml` in the git root (or working directory).
 3. **Local directory** — `.cecli/workflows/*.yml` / `*.yaml` in the git root (or working directory), loaded in sorted order.
-4. **Merged config** — the `workflows:` key in `.cecli.conf.yml`.
-5. **Global files** — `~/.cecli/workflows/*.yml` / `*.yaml`, loaded in sorted order.
+4. **Standard local directory** — `.agents/workflows/*.yml` / `*.yaml` in the git root (or working directory), loaded in sorted order.
+5. **Merged config** — the `workflows:` key in `.cecli.conf.yml`.
+6. **Global files** — `~/.cecli/workflows/*.yml` / `*.yaml`, loaded in sorted order.
 
 On a duplicate `name`, the higher-precedence definition wins and the loser is skipped with a warning. A source that fails to parse is skipped with a warning as well — a malformed workflows file never crashes startup.
 

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/cecli_dev.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cecli-dev
-Version: 1.8.1rc1
+Version: 1.8.1rc2
 Summary: we also can't pronounce cecli
 Project-URL: Homepage, https://github.com/dwash96/cecli
 Project-URL: Documentation, https://github.com/dwash96/cecli

cecli_dev-1.8.1rc2/cecli_dev.egg-info/scm_version.json

@@ -0,0 +1,8 @@
+{
+  "tag": "1.8.1rc2",
+  "distance": 0,
+  "node": "g9c5114f1f4ccf75c0b68aca1830a9232d48f1ed0",
+  "dirty": false,
+  "branch": "HEAD",
+  "node_date": "2026-10-14"
+}

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/tests/helpers/observations/test_observation_service.py

@@ -247,3 +247,48 @@ async def test_run_observation_uses_formatted_chat_chunks():
     # The observation prompt is appended as the final user message
     assert sent_messages[-1]["role"] == "user"
     assert "Observation Prompt" in sent_messages[-1]["content"]
+
+
+@pytest.mark.asyncio
+async def test_compaction_bails_out_when_fixed_context_dominates():
+    """A large file in the chat cannot be compacted away, so bail instead of looping."""
+    from cecli.coders.base_coder import Coder
+
+    coder = MagicMock(spec=Coder)
+    coder.uuid = "test-coder-fixed-context"
+    coder.enable_context_compaction = True
+    coder.context_compaction_max_tokens = 1000
+    coder.context_compaction_summary_tokens = 100
+    coder.io = MagicMock()
+    coder.gpt_prompts = MagicMock()
+    coder._rate_limit_sleep = AsyncMock()
+
+    # done/cur/diff/file_context are empty, so the entire context is the fixed
+    # context (chat files, repo map, prompts). Summarizing cannot shrink it.
+    coder.summarizer = MagicMock()
+    coder.summarizer.count_tokens.side_effect = [0, 0, 0, 0, 6000]
+    coder.summarizer.summarize_all_as_text = AsyncMock(return_value="Summary Text")
+
+    mock_conv_manager = MagicMock()
+    mock_conv_manager.get_messages_dict.return_value = []
+
+    obs = MagicMock()
+    obs.check_and_trigger = AsyncMock()
+
+    with (
+        patch(
+            "cecli.coders.base_coder.ConversationService.get_manager",
+            return_value=mock_conv_manager,
+        ),
+        patch(
+            "cecli.coders.base_coder.ObservationService.get_instance",
+            return_value=obs,
+        ),
+    ):
+        await Coder.compact_context_if_needed(coder, force=True)
+
+    # Compaction must not summarize anything...
+    coder.summarizer.summarize_all_as_text.assert_not_called()
+    # ...and must tell the user why, with a remedy.
+    coder.io.tool_warning.assert_called_once()
+    coder.io.tool_output.assert_any_call("- Use /drop to remove unneeded files from the chat")

{cecli_dev-1.8.1rc1 → cecli_dev-1.8.1rc2}/tests/helpers/workflows/test_config.py

@@ -52,6 +52,14 @@ def _local_dir_workflow(repo: Path, filename: str, text: str) -> Path:
     return path
 
 
+def _agents_dir_workflow(repo: Path, filename: str, text: str) -> Path:
+    wdir = repo / ".agents" / "workflows"
+    wdir.mkdir(parents=True, exist_ok=True)
+    path = wdir / filename
+    path.write_text(text, encoding="utf-8")
+    return path
+
+
 # ── Case 1: single-mapping file, defaults ───────────────────────────────────
 
 
@@ -512,6 +520,36 @@ def test_local_dir_wins_over_merged_config_and_home(isolated):
     assert [wf.tasks[0].prompt for wf in configs] == ["dir"]
 
 
+def test_agents_dir_files_are_loaded(isolated):
+    _agents_dir_workflow(isolated["repo"], "pipelines.yml", _mapping_text("agents-dir", "check"))
+
+    configs = load_workflows(git_root=isolated["repo"])
+
+    assert [wf.name for wf in configs] == ["agents-dir"]
+    assert configs[0].source_path == str(
+        isolated["repo"] / ".agents" / "workflows" / "pipelines.yml"
+    )
+
+
+def test_cecli_dir_wins_over_agents_dir(isolated):
+    _agents_dir_workflow(isolated["repo"], "dup.yml", _mapping_text("dup", "agents"))
+    _local_dir_workflow(isolated["repo"], "dup.yml", _mapping_text("dup", "cecli"))
+
+    configs = load_workflows(git_root=isolated["repo"])
+
+    assert [wf.tasks[0].prompt for wf in configs] == ["cecli"]
+
+
+def test_agents_dir_wins_over_merged_config_and_home(isolated):
+    _agents_dir_workflow(isolated["repo"], "dir.yml", _mapping_text("dup", "agents"))
+    _home_workflow(isolated["wdir"], "a_home.yml", _mapping_text("dup", "home"))
+    merged = {"workflows": [{"name": "dup", "tasks": [{"prompt": "merged"}]}]}
+
+    configs = load_workflows(merged_config=merged, git_root=isolated["repo"])
+
+    assert [wf.tasks[0].prompt for wf in configs] == ["agents"]
+
+
 def test_definition_hash_stable_across_key_order():
     one = models.WorkflowConfig.from_dict(
         {"name": "h", "tasks": [{"prompt": "a", "name": "t1"}], "enabled": True}

cecli_dev-1.8.1rc1/cecli_dev.egg-info/scm_version.json

@@ -1,8 +0,0 @@
-{
-  "tag": "1.8.1rc1",
-  "distance": 0,
-  "node": "g15af71fcdb6adc69eaa552bd209a8bc1838445ba",
-  "dirty": false,
-  "branch": "HEAD",
-  "node_date": "2026-10-13"
-}