zai-cli 0.1.8__tar.gz → 0.2.0__tar.gz

{zai_cli-0.1.8 → zai_cli-0.2.0}/CHANGELOG.md

@@ -5,6 +5,43 @@ All notable changes to this project are documented in this file.
 The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and the project uses [Semantic Versioning](https://semver.org/).
 
+## [0.2.0] - 2026-06-30
+
+### Added
+
+- Added project rule loading from `ZAI.md`, `.zai/rules.md`, `AGENTS.md`,
+  `CLAUDE.md`, and `GEMINI.md`.
+- Added interactive `/rules` and top-level `zai rules` to show loaded project
+  instruction files.
+- Added project-rule visibility to `zai context` and interactive `/context`.
+
+### Changed
+
+- Interactive and one-shot chat now pin loaded project rules into model context
+  so the assistant can follow repo-specific instructions.
+- Project rule loading is bounded, skips unreadable/oversized files, and redacts
+  obvious secrets before rules are added to context.
+
+## [0.1.9] - 2026-06-30
+
+### Added
+
+- Added provider-neutral token estimation helpers and context usage reporting.
+- Added `zai context tokens`, `zai context summary`, interactive
+  `/context tokens`, and interactive `/context summary`.
+- Added interactive `/compact` to summarize older session history.
+- Added `zai session compact <name-or-id>` for saved sessions.
+- Added compacted-summary metadata to saved session payloads.
+
+### Changed
+
+- Automatic context compaction now uses configurable thresholds and prints a
+  transparent message when older history is summarized.
+- Context status now shows estimated token usage, estimated context limit,
+  remaining tokens, and compacted-summary availability.
+- Long-session summaries redact obvious API keys, bearer tokens, passwords, and
+  secret values before storage/display.
+
 ## [0.1.8] - 2026-06-29
 
 ### Added

{zai_cli-0.1.8/zai_cli.egg-info → zai_cli-0.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: zai-cli
-Version: 0.1.8
+Version: 0.2.0
 Summary: Your personal AI CLI — free, fast, and smart
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/HumaizaNaz/zai_cli
@@ -343,10 +343,49 @@ When `zai` starts in a folder that already has `.zai/project_memory.json`, it
 loads a short pinned summary so future answers can use that folder's context.
 Different folders keep different memories.
 
+## Project rules
+
+Add project instruction files when you want `zai` to consistently follow repo
+rules without copy-pasting them into every prompt.
+
+Supported files:
+
+```text
+ZAI.md
+.zai/rules.md
+AGENTS.md
+CLAUDE.md
+GEMINI.md
+```
+
+When `zai` starts, it loads bounded text from these files in the project root
+and current folder, redacts obvious secrets, and pins the rules into model
+context. Use these files for coding style, preferred test commands, folders to
+avoid, release steps, and project-specific constraints.
+
+Inspect loaded rules:
+
+```text
+zai rules
+/rules
+zai context
+/context
+```
+
+Project rules are different from project memory:
+
+- Project rules are explicit instructions written by the user/repo.
+- Project memory is a generated or remembered summary in
+  `.zai/project_memory.json`.
+- Repo maps are generated structure summaries used on demand.
+- Session history is the current or saved conversation.
+
 Memory types:
 
 - Session history: previous conversation messages saved globally under `~/.zai`.
 - Project memory: small per-folder facts in `.zai/project_memory.json`.
+- Project rules: explicit instruction files such as `ZAI.md` or
+  `.zai/rules.md`.
 - Repo map: generated structure summary used for repository questions; it is not
   long-term memory.
 
@@ -393,6 +432,7 @@ zai auth login <provider>   Connect a provider API key
 zai auth logout <provider>  Remove a provider API key from zai
 zai auth doctor             Show auth troubleshooting guidance
 zai context                 Show loaded project context status
+zai rules                   Show loaded project instruction files
 zai memory show             Show last task metadata
 zai memory projects         List remembered projects
 zai memory project          Show project-scoped memory
@@ -482,6 +522,7 @@ Inside `zai` interactive mode:
 /model <name>               Change model for this interactive session
 /providers                  Show provider connection status
 /context                    Show loaded project context status
+/rules                      Show loaded project instruction files
 /memory                     Show last task metadata
 /commands                   List built-in and plugin slash commands
 ```
@@ -506,6 +547,20 @@ Use `zai context` or `/context` to inspect the current folder, whether project
 memory exists, whether a repo-map cache is available, and how many conversation
 messages are loaded in the current interactive session.
 
+For token/context usage:
+
+```text
+zai context tokens
+zai context summary
+/context tokens
+/context summary
+/compact
+```
+
+`zai context tokens` and `/context tokens` show an approximate provider-neutral
+token estimate, the configured context limit, remaining tokens, and whether the
+session is approaching the compaction threshold.
+
 Project memory is separate from current model context. It is a small
 `.zai/project_memory.json` summary created with `zai memory refresh` or
 `/memory refresh`.
@@ -514,7 +569,10 @@ Repo maps are summaries/indexes used on demand for repository questions. They
 are not full source files loaded into every model call. If a file has not been
 read or indexed for the current task, treat it as not loaded yet.
 
-Token tracking and automatic long-session compaction are planned for `0.1.9`.
+For long interactive sessions, `/compact` summarizes older conversation and
+tool history so the active session can continue with less context pressure. The
+summary preserves the goal, decisions, recent work, commands/results where
+visible, and pending context in a compact form.
 
 All chat, workflow, utility, and agent requests use a model-aware context
 pipeline:
@@ -528,8 +586,9 @@ pipeline:
 - context usage uses a conservative shared token estimate, while provider
   response usage is retained where the SDK reports it.
 
-This keeps requests bounded, but it is not a replacement for future
-long-session token tracking and automatic compaction.
+This keeps requests bounded. Compaction is still approximate, so save important
+stable project facts explicitly with `/remember <fact>` or `zai memory remember
+<fact>`.
 
 ## Providers and fallback
 

{zai_cli-0.1.8 → zai_cli-0.2.0}/README.md

@@ -286,10 +286,49 @@ When `zai` starts in a folder that already has `.zai/project_memory.json`, it
 loads a short pinned summary so future answers can use that folder's context.
 Different folders keep different memories.
 
+## Project rules
+
+Add project instruction files when you want `zai` to consistently follow repo
+rules without copy-pasting them into every prompt.
+
+Supported files:
+
+```text
+ZAI.md
+.zai/rules.md
+AGENTS.md
+CLAUDE.md
+GEMINI.md
+```
+
+When `zai` starts, it loads bounded text from these files in the project root
+and current folder, redacts obvious secrets, and pins the rules into model
+context. Use these files for coding style, preferred test commands, folders to
+avoid, release steps, and project-specific constraints.
+
+Inspect loaded rules:
+
+```text
+zai rules
+/rules
+zai context
+/context
+```
+
+Project rules are different from project memory:
+
+- Project rules are explicit instructions written by the user/repo.
+- Project memory is a generated or remembered summary in
+  `.zai/project_memory.json`.
+- Repo maps are generated structure summaries used on demand.
+- Session history is the current or saved conversation.
+
 Memory types:
 
 - Session history: previous conversation messages saved globally under `~/.zai`.
 - Project memory: small per-folder facts in `.zai/project_memory.json`.
+- Project rules: explicit instruction files such as `ZAI.md` or
+  `.zai/rules.md`.
 - Repo map: generated structure summary used for repository questions; it is not
   long-term memory.
 
@@ -336,6 +375,7 @@ zai auth login <provider>   Connect a provider API key
 zai auth logout <provider>  Remove a provider API key from zai
 zai auth doctor             Show auth troubleshooting guidance
 zai context                 Show loaded project context status
+zai rules                   Show loaded project instruction files
 zai memory show             Show last task metadata
 zai memory projects         List remembered projects
 zai memory project          Show project-scoped memory
@@ -425,6 +465,7 @@ Inside `zai` interactive mode:
 /model <name>               Change model for this interactive session
 /providers                  Show provider connection status
 /context                    Show loaded project context status
+/rules                      Show loaded project instruction files
 /memory                     Show last task metadata
 /commands                   List built-in and plugin slash commands
 ```
@@ -449,6 +490,20 @@ Use `zai context` or `/context` to inspect the current folder, whether project
 memory exists, whether a repo-map cache is available, and how many conversation
 messages are loaded in the current interactive session.
 
+For token/context usage:
+
+```text
+zai context tokens
+zai context summary
+/context tokens
+/context summary
+/compact
+```
+
+`zai context tokens` and `/context tokens` show an approximate provider-neutral
+token estimate, the configured context limit, remaining tokens, and whether the
+session is approaching the compaction threshold.
+
 Project memory is separate from current model context. It is a small
 `.zai/project_memory.json` summary created with `zai memory refresh` or
 `/memory refresh`.
@@ -457,7 +512,10 @@ Repo maps are summaries/indexes used on demand for repository questions. They
 are not full source files loaded into every model call. If a file has not been
 read or indexed for the current task, treat it as not loaded yet.
 
-Token tracking and automatic long-session compaction are planned for `0.1.9`.
+For long interactive sessions, `/compact` summarizes older conversation and
+tool history so the active session can continue with less context pressure. The
+summary preserves the goal, decisions, recent work, commands/results where
+visible, and pending context in a compact form.
 
 All chat, workflow, utility, and agent requests use a model-aware context
 pipeline:
@@ -471,8 +529,9 @@ pipeline:
 - context usage uses a conservative shared token estimate, while provider
   response usage is retained where the SDK reports it.
 
-This keeps requests bounded, but it is not a replacement for future
-long-session token tracking and automatic compaction.
+This keeps requests bounded. Compaction is still approximate, so save important
+stable project facts explicitly with `/remember <fact>` or `zai memory remember
+<fact>`.
 
 ## Providers and fallback
 

{zai_cli-0.1.8 → zai_cli-0.2.0}/tests/test_interactive.py

@@ -180,6 +180,71 @@ def test_interactive_session_rename_routes_arguments(tmp_path, monkeypatch):
 def test_interactive_project_memory_commands(tmp_path, monkeypatch, capsys):
     monkeypatch.chdir(tmp_path)
     (tmp_path / "pyproject.toml").write_text("[project]\nname='demo'\n")
+    with patch("zai.cli.interactive.has_available_provider", return_value=True):
+        with patch("zai.cli.interactive.fire_hook", return_value=True):
+            with patch("zai.cli.interactive.add_project"):
+                with patch("zai.cli.interactive.plugin_loader.load_all", return_value={}):
+                    with patch("zai.cli.interactive.plugin_loader.get_errors", return_value={}):
+                        with patch("zai.cli.interactive._connect_mcp_servers"):
+                            with patch(
+                                "zai.cli.interactive.InteractiveInput.prompt",
+                                side_effect=[
+                                    "/memory project",
+                                    "/memory refresh",
+                                    "/remember uses pytest",
+                                    "/forget 1",
+                                    EOFError(),
+                                ],
+                            ):
+                                with patch("zai.cli.interactive.run_agent") as run_agent:
+                                    run_interactive("groq")
+
+    output = capsys.readouterr().out
+    assert "Project Memory" in output
+    assert "Project Memory Refreshed" in output
+    assert "Remembered" in output
+    assert "Forgot" in output
+    run_agent.assert_not_called()
+
+
+def test_interactive_auth_commands_do_not_call_ai(tmp_path, monkeypatch, capsys):
+    monkeypatch.chdir(tmp_path)
+    with patch("zai.cli.interactive.has_available_provider", return_value=True):
+        with patch("zai.cli.interactive.fire_hook", return_value=True):
+            with patch("zai.cli.interactive.add_project"):
+                with patch("zai.cli.interactive.plugin_loader.load_all", return_value={}):
+                    with patch("zai.cli.interactive.plugin_loader.get_errors", return_value={}):
+                        with patch("zai.cli.interactive._connect_mcp_servers"):
+                            with patch(
+                                "zai.cli.interactive.InteractiveInput.prompt",
+                                side_effect=[
+                                    "/auth status",
+                                    "/auth login gemini",
+                                    "/auth logout gemini",
+                                    EOFError(),
+                                ],
+                            ):
+                                with (
+                                    patch("zai.cli.interactive.Prompt.ask", return_value="gemini-key"),
+                                    patch("zai.cli.interactive.Confirm.ask", return_value=False),
+                                    patch("zai.cli.interactive.save_provider_key") as save,
+                                    patch("zai.cli.interactive.delete_provider_key", return_value=True) as delete,
+                                    patch("zai.cli.interactive.run_agent") as run_agent,
+                                ):
+                                    run_interactive("groq")
+
+    output = capsys.readouterr().out
+    assert "Provider status" in output
+    assert "Saved Gemini API key" in output
+    save.assert_called_once_with("gemini", "gemini-key")
+    delete.assert_called_once_with("gemini")
+    run_agent.assert_not_called()
+
+
+def test_interactive_context_and_providers_do_not_call_ai(tmp_path, monkeypatch, capsys):
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "app.py").write_text("print('hi')\n")
+    (tmp_path / "ZAI.md").write_text("Use pytest.\n")
     with patch("zai.cli.interactive.has_available_provider", return_value=True):
         with patch("zai.cli.interactive.fire_hook", return_value=True):
             with patch("zai.cli.interactive.plugin_loader.load_all", return_value={}):
@@ -188,10 +253,11 @@ def test_interactive_project_memory_commands(tmp_path, monkeypatch, capsys):
                         with patch(
                             "zai.cli.interactive.InteractiveInput.prompt",
                             side_effect=[
-                                "/memory project",
-                                "/memory refresh",
-                                "/remember uses pytest",
-                                "/forget 1",
+                                "/providers",
+                                "/context",
+                                "/rules",
+                                "/context tokens",
+                                "/context summary",
                                 EOFError(),
                             ],
                         ):
@@ -199,15 +265,21 @@ def test_interactive_project_memory_commands(tmp_path, monkeypatch, capsys):
                                 run_interactive("groq")
 
     output = capsys.readouterr().out
-    assert "Project Memory" in output
-    assert "Project Memory Refreshed" in output
-    assert "Remembered" in output
-    assert "Forgot" in output
+    assert "Provider status" in output
+    assert "Project context" in output
+    assert "Project rules loaded:" in output
+    assert "ZAI.md" in output
+    assert "Project Rules" in output
+    assert "unread files are not loaded" in output
+    assert "Estimated tokens used" in output
+    assert "No compacted summary" in output
+    assert "app.py is loaded" not in output
     run_agent.assert_not_called()
 
 
-def test_interactive_auth_commands_do_not_call_ai(tmp_path, monkeypatch, capsys):
+def test_interactive_pins_project_rules_before_agent_call(tmp_path, monkeypatch):
     monkeypatch.chdir(tmp_path)
+    (tmp_path / "ZAI.md").write_text("Use the custom test command.\n")
     with patch("zai.cli.interactive.has_available_provider", return_value=True):
         with patch("zai.cli.interactive.fire_hook", return_value=True):
             with patch("zai.cli.interactive.plugin_loader.load_all", return_value={}):
@@ -215,33 +287,36 @@ def test_interactive_auth_commands_do_not_call_ai(tmp_path, monkeypatch, capsys)
                     with patch("zai.cli.interactive._connect_mcp_servers"):
                         with patch(
                             "zai.cli.interactive.InteractiveInput.prompt",
-                            side_effect=[
-                                "/auth status",
-                                "/auth login gemini",
-                                "/auth logout gemini",
-                                EOFError(),
-                            ],
+                            side_effect=["do work", EOFError()],
                         ):
                             with (
-                                patch("zai.cli.interactive.Prompt.ask", return_value="gemini-key"),
+                                patch(
+                                    "zai.cli.interactive.run_agent",
+                                    return_value="done",
+                                ) as run_agent,
                                 patch("zai.cli.interactive.Confirm.ask", return_value=False),
-                                patch("zai.cli.interactive.save_provider_key") as save,
-                                patch("zai.cli.interactive.delete_provider_key", return_value=True) as delete,
-                                patch("zai.cli.interactive.run_agent") as run_agent,
                             ):
                                 run_interactive("groq")
 
-    output = capsys.readouterr().out
-    assert "Provider status" in output
-    assert "Saved Gemini API key" in output
-    save.assert_called_once_with("gemini", "gemini-key")
-    delete.assert_called_once_with("gemini")
-    run_agent.assert_not_called()
+    history = run_agent.call_args.args[1]
+    assert any(
+        message.pinned and "Use the custom test command" in message.content
+        for message in history
+    )
 
 
-def test_interactive_context_and_providers_do_not_call_ai(tmp_path, monkeypatch, capsys):
+def test_interactive_manual_compact_summarizes_history(tmp_path, monkeypatch, capsys):
     monkeypatch.chdir(tmp_path)
-    (tmp_path / "app.py").write_text("print('hi')\n")
+    prompts = [
+        "task one",
+        "task two",
+        "task three",
+        "task four",
+        "task five",
+        "/compact",
+        "/context summary",
+        EOFError(),
+    ]
     with patch("zai.cli.interactive.has_available_provider", return_value=True):
         with patch("zai.cli.interactive.fire_hook", return_value=True):
             with patch("zai.cli.interactive.plugin_loader.load_all", return_value={}):
@@ -249,17 +324,30 @@ def test_interactive_context_and_providers_do_not_call_ai(tmp_path, monkeypatch,
                     with patch("zai.cli.interactive._connect_mcp_servers"):
                         with patch(
                             "zai.cli.interactive.InteractiveInput.prompt",
-                            side_effect=["/providers", "/context", EOFError()],
+                            side_effect=prompts,
                         ):
-                            with patch("zai.cli.interactive.run_agent") as run_agent:
+                            with (
+                                patch(
+                                    "zai.cli.interactive.run_agent",
+                                    side_effect=[
+                                        "answer one",
+                                        "answer two",
+                                        "answer three",
+                                        "answer four",
+                                        "answer five",
+                                    ],
+                                ),
+                                patch(
+                                    "zai.cli.interactive.Confirm.ask",
+                                    return_value=False,
+                                ),
+                            ):
                                 run_interactive("groq")
 
     output = capsys.readouterr().out
-    assert "Provider status" in output
-    assert "Project context" in output
-    assert "unread files are not loaded" in output
-    assert "app.py is loaded" not in output
-    run_agent.assert_not_called()
+    assert "Context compacted" in output
+    assert "Current compacted summary" in output
+    assert "task one" in output
 
 
 def test_interactive_offers_project_memory_after_agent_task(tmp_path, monkeypatch, capsys):

zai_cli-0.2.0/tests/test_project_rules.py

@@ -0,0 +1,33 @@
+from zai.core.project_rules import (
+    format_project_rules,
+    load_project_rules,
+    project_rules_prompt,
+)
+
+
+def test_load_project_rules_from_root_and_current_folder(tmp_path):
+    (tmp_path / ".git").mkdir()
+    (tmp_path / "ZAI.md").write_text("Use pytest.\n", encoding="utf-8")
+    nested = tmp_path / "pkg"
+    nested.mkdir()
+    (nested / "AGENTS.md").write_text("Nested rule.\n", encoding="utf-8")
+
+    rules = load_project_rules(nested)
+
+    assert [rule.relative_path for rule in rules] == ["ZAI.md", "pkg/AGENTS.md"]
+    assert "Use pytest" in project_rules_prompt(rules)
+    assert "pkg/AGENTS.md" in format_project_rules(rules)
+
+
+def test_project_rules_redact_secrets_and_skip_large_files(tmp_path):
+    (tmp_path / "ZAI.md").write_text(
+        "api_key = sk-secret-value-1234567890\nFollow style.\n",
+        encoding="utf-8",
+    )
+    (tmp_path / "CLAUDE.md").write_text("x" * (70 * 1024), encoding="utf-8")
+
+    rules = load_project_rules(tmp_path)
+
+    assert len(rules) == 1
+    assert "sk-secret" not in rules[0].content
+    assert "[REDACTED_SECRET]" in rules[0].content

{zai_cli-0.1.8 → zai_cli-0.2.0}/tests/test_settings_cli.py

@@ -141,17 +141,50 @@ def test_context_command_is_honest_about_unloaded_files(tmp_path, monkeypatch):
     app = settings_app()
     monkeypatch.chdir(tmp_path)
     (tmp_path / "app.py").write_text("print('hi')\n")
+    (tmp_path / "ZAI.md").write_text("Use pytest.\n")
 
     result = CliRunner().invoke(app, ["context"])
 
     assert result.exit_code == 0
     assert f"Folder: {tmp_path}" in result.output
     assert "Project memory: missing" in result.output
+    assert "Project rules:" in result.output
+    assert "ZAI.md" in result.output
     assert "Repo map/index: not loaded yet" in result.output
     assert "unread files are not loaded" in result.output
+    assert "Estimated tokens used" in result.output
     assert "app.py is loaded" not in result.output
 
 
+def test_rules_command_shows_loaded_project_rules(tmp_path, monkeypatch):
+    app = settings_app()
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / ".zai").mkdir()
+    (tmp_path / ".zai" / "rules.md").write_text("Never commit secrets.\n")
+
+    result = CliRunner().invoke(app, ["rules"])
+
+    assert result.exit_code == 0
+    assert "Project Rules" in result.output
+    assert ".zai/rules.md" in result.output
+
+
+def test_context_tokens_and_summary_commands(tmp_path, monkeypatch):
+    app = settings_app()
+    monkeypatch.chdir(tmp_path)
+
+    tokens = CliRunner().invoke(app, ["context", "tokens"])
+    summary = CliRunner().invoke(app, ["context", "summary"])
+    bad = CliRunner().invoke(app, ["context", "bad"])
+
+    assert tokens.exit_code == 0
+    assert "Estimated tokens used" in tokens.output
+    assert "Compacted summary" in tokens.output
+    assert summary.exit_code == 0
+    assert "No compacted summary" in summary.output
+    assert "Usage: zai context" in bad.output
+
+
 def test_model_check_reports_valid_and_invalid_configuration():
     app = settings_app()
     runner = CliRunner()
@@ -408,6 +441,27 @@ def test_session_rename_delete_and_export():
     assert "Export failed" in export_failed.output
 
 
+def test_session_compact_routes_history():
+    from zai.providers.base import Message
+
+    app = settings_app()
+    history = [
+        Message(role="user", content=f"message {index}")
+        for index in range(12)
+    ]
+    info = {"name": "auth", "project_path": "project"}
+    with (
+        patch("zai.cli.settings.load_history", return_value=history),
+        patch("zai.cli.settings.get_session_info", return_value=info),
+        patch("zai.cli.settings.save_history") as save,
+    ):
+        result = CliRunner().invoke(app, ["session", "compact", "auth"])
+
+    assert result.exit_code == 0
+    assert "Session compacted" in result.output
+    save.assert_called_once()
+
+
 def test_undo_list_empty_and_populated():
     app = settings_app()
     runner = CliRunner()

zai_cli-0.2.0/tests/test_tokens.py

@@ -0,0 +1,57 @@
+from zai.core.compaction import manual_compact_messages
+from zai.core.tokens import (
+    compacted_summaries,
+    context_usage,
+    estimate_messages_tokens,
+    estimate_tokens,
+    redact_secrets,
+)
+from zai.providers.base import Message
+
+
+def test_estimate_tokens_is_stable_and_nonzero():
+    assert estimate_tokens("") == 0
+    assert estimate_tokens("hello world") >= 2
+    assert estimate_tokens("x" * 120) >= 40
+
+
+def test_context_usage_reports_thresholds():
+    messages = [Message(role="user", content="x" * 900)]
+    usage = context_usage(messages, context_limit=100)
+
+    assert usage.tokens_used == estimate_messages_tokens(messages)
+    assert usage.tokens_remaining == 0
+    assert usage.should_warn is True
+    assert usage.should_compact is True
+
+
+def test_secret_redaction_covers_common_patterns():
+    text = (
+        "OPENAI_API_KEY=sk-abcdefghijklmnopqrstuvwxyz "
+        "Authorization: Bearer abcdefghijklmnopqrstuvwxyz "
+        "password=hunter2"
+    )
+    redacted = redact_secrets(text)
+
+    assert "sk-abcdefghijklmnopqrstuvwxyz" not in redacted
+    assert "abcdefghijklmnopqrstuvwxyz" not in redacted
+    assert "hunter2" not in redacted
+    assert "[REDACTED_SECRET]" in redacted
+
+
+def test_manual_compaction_creates_summary_and_keeps_recent_messages():
+    messages = [
+        Message(role="user", content="initial goal"),
+        Message(role="assistant", content="decision made"),
+        Message(role="tool", content="api_key=sk-secretsecretsecret", tool_name="run"),
+    ]
+    messages.extend(Message(role="user", content=f"step {index}") for index in range(12))
+
+    compacted, summary = manual_compact_messages(messages, max_tokens=128_000, keep_recent=4)
+
+    assert summary is not None
+    assert "initial goal" in summary.content
+    assert "decision made" in summary.content
+    assert "sk-secretsecretsecret" not in summary.content
+    assert compacted[-1].content == "step 11"
+    assert compacted_summaries(compacted)

zai_cli-0.2.0/zai/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.2.0"