loopx 0.4.9__tar.gz → 0.5.0__tar.gz

{loopx-0.4.9/loopx.egg-info → loopx-0.5.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: loopx
-Version: 0.4.9
+Version: 0.5.0
 Summary: A lightweight Loop Engineering control plane for long-running agent goals.
 Author: LoopX contributors
 License-Expression: Apache-2.0
@@ -18,7 +18,7 @@ Provides-Extra: deepseek-harness
 Requires-Dist: deepseek-harness-sdk>=0.1.0rc5; extra == "deepseek-harness"
 Provides-Extra: test
 Requires-Dist: jsonschema<5,>=4.23; extra == "test"
-Requires-Dist: mcp==1.27.2; extra == "test"
+Requires-Dist: mcp==1.28.1; extra == "test"
 Requires-Dist: types-jsonschema<5,>=4.23; extra == "test"
 Requires-Dist: pytest<9,>=8; extra == "test"
 Requires-Dist: pytest-cov<7,>=5; extra == "test"
@@ -66,6 +66,46 @@ to hand off across turns, tools, and agents.
 - **Getting started** - install, connect a project, and run your first governed loop. [Guide](docs/guides/getting-started.md)
 - **Docs** - the full reference and operations site. [LoopX Docs](https://huangruiteng.github.io/loopx/docs/)
 
+## Meet the Personal Agent Workspace
+
+Turn scattered agent sessions into one local-first workspace for Goals,
+attention, conversations, tasks, files, schedules, and recovery.
+
+<a href="docs/assets/personal-workspace/loopx-dashboard-launch.mp4">
+  <img src="docs/assets/personal-workspace/loopx-dashboard-tour.webp" alt="Animated tour of the LoopX personal Agent workspace, from one-command launch to Manager overview, Goal task board, protected action preview, and Goal details" width="960">
+</a>
+
+The workspace gives you one place to:
+
+- see what needs you, what is running, what is being watched, and what is
+  scheduled or stopped;
+- continue across Codex, Claude Code, direct-model, and other registered Agent
+  sessions without losing Goal state or evidence;
+- review protected changes through typed preview, explicit confirmation, and
+  receipts while LoopX state—not the browser—remains authoritative.
+
+```bash
+loopx dashboard
+```
+
+`loopx dashboard` is the supported browser/PWA launch path. For a native
+window, the experimental source-built Tauri shell reuses the same loopback
+status and Chat services without becoming another state authority:
+
+```bash
+cd apps/desktop/loopx-control-plane
+npm install
+npm run dev
+```
+
+[Read the desktop shell guide](apps/desktop/loopx-control-plane/README.md).
+Closing the window stops only the service processes that the shell started;
+existing LoopX services and durable Goal state remain intact.
+
+[Watch the full 32-second walkthrough](docs/assets/personal-workspace/loopx-dashboard-launch.mp4)
+· [Read the workspace guide](docs/guides/personal-workspace-user-guide.md)
+· [Try the five-minute tour](docs/guides/personal-workspace-trial-guide.md)
+
 <a id="how-it-works"></a>
 
 ## Why LoopX
@@ -370,6 +410,7 @@ evidence → recovery; continuation → governance.
 | Quota and interaction contract | Decides whether a turn should deliver, ask, wait, self-repair, or stay quiet. | `loopx quota should-run`, [quota allocation](docs/quota-allocation.md) |
 | Agent runtime bridges | Keeps Codex App, Codex CLI, Claude Code, and generic workers aligned with the same guard. | `loopx heartbeat-prompt`, `loopx codex-cli-bootstrap-message`, `loopx worker-bridge` |
 | Operator surfaces | Renders compact status without making the browser the state authority. | `loopx serve-status`, [dashboard](apps/presentation/dashboard/README.md) |
+| Session dash | Starts a live single-page panel that tracks fleet progress: sessions, their goals, and each goal's status/todo progress, with result statistics; auto-refreshes in place. | `loopx dash`, [session dash design](docs/product/surfaces/session-dash-panel-design.md) |
 | External projections | Projects todos and gates into collaboration surfaces while LoopX remains authoritative. | `loopx lark-kanban`, [Lark Kanban adapter](docs/integrations/lark-kanban-control-plane-adapter.md) |
 | Domain capabilities | Packages repeatable work lanes such as issue fixing, content operations, value connector planning, ML experiment advice, benchmark evidence, and Explore. | `loopx issue-fix`, `loopx content-ops`, `loopx value-connectors`, `loopx ml-experiment`, `loopx benchmark`, [Explore](loopx/capabilities/explore/README.md) |
 | Experimental context learning | Lets named registered agents trial provider-neutral Reward Memory through ignored, default-off project configuration. OpenViking is one provider option, not a global dependency. | `loopx reward-memory experiment-status`, [Reward Memory architecture](loopx/capabilities/reward_memory/README.md) |

{loopx-0.4.9 → loopx-0.5.0}/README.md

@@ -37,6 +37,46 @@ to hand off across turns, tools, and agents.
 - **Getting started** - install, connect a project, and run your first governed loop. [Guide](docs/guides/getting-started.md)
 - **Docs** - the full reference and operations site. [LoopX Docs](https://huangruiteng.github.io/loopx/docs/)
 
+## Meet the Personal Agent Workspace
+
+Turn scattered agent sessions into one local-first workspace for Goals,
+attention, conversations, tasks, files, schedules, and recovery.
+
+<a href="docs/assets/personal-workspace/loopx-dashboard-launch.mp4">
+  <img src="docs/assets/personal-workspace/loopx-dashboard-tour.webp" alt="Animated tour of the LoopX personal Agent workspace, from one-command launch to Manager overview, Goal task board, protected action preview, and Goal details" width="960">
+</a>
+
+The workspace gives you one place to:
+
+- see what needs you, what is running, what is being watched, and what is
+  scheduled or stopped;
+- continue across Codex, Claude Code, direct-model, and other registered Agent
+  sessions without losing Goal state or evidence;
+- review protected changes through typed preview, explicit confirmation, and
+  receipts while LoopX state—not the browser—remains authoritative.
+
+```bash
+loopx dashboard
+```
+
+`loopx dashboard` is the supported browser/PWA launch path. For a native
+window, the experimental source-built Tauri shell reuses the same loopback
+status and Chat services without becoming another state authority:
+
+```bash
+cd apps/desktop/loopx-control-plane
+npm install
+npm run dev
+```
+
+[Read the desktop shell guide](apps/desktop/loopx-control-plane/README.md).
+Closing the window stops only the service processes that the shell started;
+existing LoopX services and durable Goal state remain intact.
+
+[Watch the full 32-second walkthrough](docs/assets/personal-workspace/loopx-dashboard-launch.mp4)
+· [Read the workspace guide](docs/guides/personal-workspace-user-guide.md)
+· [Try the five-minute tour](docs/guides/personal-workspace-trial-guide.md)
+
 <a id="how-it-works"></a>
 
 ## Why LoopX
@@ -341,6 +381,7 @@ evidence → recovery; continuation → governance.
 | Quota and interaction contract | Decides whether a turn should deliver, ask, wait, self-repair, or stay quiet. | `loopx quota should-run`, [quota allocation](docs/quota-allocation.md) |
 | Agent runtime bridges | Keeps Codex App, Codex CLI, Claude Code, and generic workers aligned with the same guard. | `loopx heartbeat-prompt`, `loopx codex-cli-bootstrap-message`, `loopx worker-bridge` |
 | Operator surfaces | Renders compact status without making the browser the state authority. | `loopx serve-status`, [dashboard](apps/presentation/dashboard/README.md) |
+| Session dash | Starts a live single-page panel that tracks fleet progress: sessions, their goals, and each goal's status/todo progress, with result statistics; auto-refreshes in place. | `loopx dash`, [session dash design](docs/product/surfaces/session-dash-panel-design.md) |
 | External projections | Projects todos and gates into collaboration surfaces while LoopX remains authoritative. | `loopx lark-kanban`, [Lark Kanban adapter](docs/integrations/lark-kanban-control-plane-adapter.md) |
 | Domain capabilities | Packages repeatable work lanes such as issue fixing, content operations, value connector planning, ML experiment advice, benchmark evidence, and Explore. | `loopx issue-fix`, `loopx content-ops`, `loopx value-connectors`, `loopx ml-experiment`, `loopx benchmark`, [Explore](loopx/capabilities/explore/README.md) |
 | Experimental context learning | Lets named registered agents trial provider-neutral Reward Memory through ignored, default-off project configuration. OpenViking is one provider option, not a global dependency. | `loopx reward-memory experiment-status`, [Reward Memory architecture](loopx/capabilities/reward_memory/README.md) |

{loopx-0.4.9 → loopx-0.5.0}/loopx/__init__.py

@@ -2,4 +2,4 @@
 
 __all__ = ["__version__"]
 
-__version__ = "0.4.9"
+__version__ = "0.5.0"

{loopx-0.4.9 → loopx-0.5.0}/loopx/bootstrap.py

@@ -307,9 +307,14 @@ def onboarding_next_action(
     accept_onboarding_agent_todos: bool,
     begin_autonomous_advance: bool,
     codex_app_heartbeat: str,
+    include_connection_validation: bool = True,
 ) -> str:
     if not onboarding_scan:
-        validation_action = onboarding_connection_validation_action(adapter_kind)
+        validation_action = (
+            onboarding_connection_validation_action(adapter_kind)
+            if include_connection_validation
+            else None
+        )
         if validation_action:
             return validation_action["text"]
         return "Initial routing is owned by the connected domain adapter."
@@ -361,10 +366,15 @@ def apply_onboarding_todos_to_state(
     accept_onboarding_agent_todos: bool,
     begin_autonomous_advance: bool,
     codex_app_heartbeat: str,
+    include_connection_validation: bool = True,
 ) -> str:
     if not onboarding_scan:
         lines = text.splitlines()
-        action = onboarding_connection_validation_action(adapter_kind)
+        action = (
+            onboarding_connection_validation_action(adapter_kind)
+            if include_connection_validation
+            else None
+        )
         if action:
             add_todo_to_lines(
                 lines,
@@ -437,6 +447,7 @@ def render_state_markdown(
     accept_onboarding_agent_todos: bool = False,
     begin_autonomous_advance: bool = False,
     codex_app_heartbeat: str = "ask",
+    include_connection_validation: bool = True,
     handoff_mode: str = HANDOFF_MODE_LEGACY,
 ) -> str:
     safe_objective = objective.replace('"', '\\"')
@@ -454,6 +465,7 @@ def render_state_markdown(
         accept_onboarding_agent_todos=accept_onboarding_agent_todos,
         begin_autonomous_advance=begin_autonomous_advance,
         codex_app_heartbeat=codex_app_heartbeat,
+        include_connection_validation=include_connection_validation,
     )
     # ``handoff_mode`` travels in the state front matter (RFC shared-goal
     # authority, Appendix B). Legacy is the absent default and is never
@@ -524,6 +536,7 @@ adapter_id: {goal_id}
         accept_onboarding_agent_todos=accept_onboarding_agent_todos,
         begin_autonomous_advance=begin_autonomous_advance,
         codex_app_heartbeat=codex_app_heartbeat,
+        include_connection_validation=include_connection_validation,
     )
 
 
@@ -565,6 +578,7 @@ def build_goal_entry(
     allowed_domains: list[str],
     write_scope: list[str],
     execution_profile: dict[str, Any] | None,
+    display_name: str | None = None,
 ) -> dict[str, Any]:
     authority_sources = []
     if goal_doc:
@@ -577,6 +591,7 @@ def build_goal_entry(
         )
     return {
         "id": goal_id,
+        **({"display_name": display_name} if display_name else {}),
         "domain": domain,
         "status": "active",
         "role": role,
@@ -676,6 +691,8 @@ def bootstrap_project(
     onboarding_max_status_paths: int = 12,
     onboarding_max_top_level_files: int = 24,
     preserve_todos: bool = False,
+    display_name: str | None = None,
+    include_connection_validation: bool = True,
     force: bool,
     dry_run: bool,
     sync_global: bool,
@@ -738,6 +755,7 @@ def bootstrap_project(
         allowed_domains=allowed_domains or [],
         write_scope=write_scope or [],
         execution_profile=execution_profile,
+        display_name=display_name,
     )
     registry, registry_goal_action = merge_goal(registry, goal_entry, force=force)
 
@@ -914,6 +932,7 @@ def bootstrap_project(
                     accept_onboarding_agent_todos=accept_onboarding_agent_todos,
                     begin_autonomous_advance=begin_autonomous_advance,
                     codex_app_heartbeat=codex_app_heartbeat,
+                    include_connection_validation=include_connection_validation,
                     handoff_mode=declared_handoff_mode,
                 ),
                 encoding="utf-8",

{loopx-0.4.9 → loopx-0.5.0}/loopx/capabilities/benchmark_toolkit/README.md

@@ -309,14 +309,18 @@ receipt.
 
 A countable experiment uses the toolkit in this order:
 
-1. Declare a `run_permission_policy_v0` and preflight the runner boundary.
-2. Launch one frozen case/arm; do not expose evaluator sources or official feedback.
-3. Capture ATIF tool evidence and a runner-owned runtime attestation.
-4. Run `integrity-qualification`; stop on any blocker.
-5. Run the independent verifier only after the agent phase.
-6. Reduce the official result through the benchmark-owned scoring path.
-7. Apply attempt-countability, treatment-fidelity, and matched-pair gates before any
-   comparison claim or ledger update.
+1. Read the project experiment board before selecting or launching another case.
+2. Declare a `run_permission_policy_v0` and preflight the runner boundary.
+3. Upsert the planned or running row, then launch one frozen case/arm; do not expose
+   evaluator sources or official feedback.
+4. Capture ATIF tool evidence and a runner-owned runtime attestation.
+5. Run `integrity-qualification`; stop on any blocker.
+6. Run the independent verifier only after the agent phase.
+7. Reduce the official result through the benchmark-owned scoring path.
+8. Upsert terminal score, countability, effort, treatment fidelity, and insight
+   status, then read the matched-comparison projection.
+9. Apply attempt-countability, treatment-fidelity, and matched-pair gates before any
+   comparison claim.
 
 Integrity qualification is necessary but not sufficient for a score claim. It does
 not establish task correctness, official score authority, experiment parity, or a
@@ -324,6 +328,55 @@ LoopX advantage. `score_claim_eligible=true` only permits the official score and
 matched-pair gates to run; `score_claim_countable` and `matched_pair_countable` stay
 false in this receipt. Those remain separate verifier and comparison contracts.
 
+## Experiment board
+
+The project-local experiment board keeps baseline, standard control or treatment,
+and diagnostic explore runs in one compact projection. It is not a second score
+authority and never stores raw task text, trajectories, logs, hidden evaluation,
+verifier output, credentials, or local paths.
+
+An agent using this capability should start or resume a study by reading the board:
+
+```bash
+loopx benchmark experiment-board-show \
+  --goal-id <goal-id> \
+  --format json
+```
+
+Preview and then execute an idempotent row update when a run starts or reaches a
+terminal state:
+
+```bash
+loopx benchmark experiment-board-upsert \
+  --goal-id <goal-id> \
+  --row-json <compact-row.json> \
+  --format json
+
+loopx benchmark experiment-board-upsert \
+  --goal-id <goal-id> \
+  --row-json <compact-row.json> \
+  --execute \
+  --format json
+```
+
+The default ledger is locked, atomically updated, and keyed by benchmark, study,
+case, and run identity. A compact row carries arm role, exact and comparison
+protocol ids, model, score metrics, countability, treatment fidelity, bounded
+effort, and an optional insight status or public-safe handle. Unknown fields and
+path-like references fail closed.
+
+Every non-baseline row names an exact `comparison_anchor_run_id`. Standard control
+or treatment rows anchor to a baseline. Explore rows use `diagnostic_only` claim
+scope and may anchor to the baseline or fixed standard arm they are examining.
+Matched comparisons require compatible benchmark, study, case, model, primary
+metric, comparison protocol, score countability, and treatment fidelity. Exact
+protocol revisions remain visible as a warning even when a declared comparison
+protocol says an older credible score remains semantically comparable.
+
+Full post-run analysis stays in private `benchmark_case_insight_v0` storage. The
+board records only its compact status or handle, so reading the board cannot widen
+the solving agent's evidence boundary.
+
 ## Post-run case insight monitor
 
 Benchmark startup should create one `continuous_monitor` todo. Whenever a case

{loopx-0.4.9 → loopx-0.5.0}/loopx/capabilities/benchmark_toolkit/__init__.py

@@ -15,6 +15,19 @@ from .container_binding import (
     compact_docker_container_binding_receipt,
     select_exact_docker_container,
 )
+from .experiment_board import (
+    BENCHMARK_EXPERIMENT_BOARD_LEDGER_FILENAME,
+    BENCHMARK_EXPERIMENT_BOARD_ROW_SCHEMA_VERSION,
+    BENCHMARK_EXPERIMENT_BOARD_SCHEMA_VERSION,
+    benchmark_experiment_board_row_key,
+    build_benchmark_experiment_board,
+    default_benchmark_experiment_board_path,
+    normalize_benchmark_experiment_board_row,
+    preview_benchmark_experiment_board_upsert,
+    read_benchmark_experiment_board_rows,
+    render_benchmark_experiment_board_markdown,
+    upsert_benchmark_experiment_board_row,
+)
 from .integrity import (
     BENCHMARK_INTEGRITY_POLICY_SCHEMA_VERSION,
     BENCHMARK_INTEGRITY_QUALIFICATION_SCHEMA_VERSION,
@@ -86,6 +99,9 @@ from .run_permissions import (
 __all__ = [
     "BENCHMARK_CANDIDATE_SOURCE_BOUNDARY_SCHEMA_VERSION",
     "BENCHMARK_EXACT_CONTAINER_BINDING_SCHEMA_VERSION",
+    "BENCHMARK_EXPERIMENT_BOARD_LEDGER_FILENAME",
+    "BENCHMARK_EXPERIMENT_BOARD_ROW_SCHEMA_VERSION",
+    "BENCHMARK_EXPERIMENT_BOARD_SCHEMA_VERSION",
     "BENCHMARK_INTEGRITY_POLICY_SCHEMA_VERSION",
     "BENCHMARK_INTEGRITY_QUALIFICATION_SCHEMA_VERSION",
     "BENCHMARK_RUNTIME_INTEGRITY_ATTESTATION_SCHEMA_VERSION",
@@ -119,7 +135,9 @@ __all__ = [
     "PublicTrajectorySummaryError",
     "StdioNativeGoalTransport",
     "attach_native_goal",
+    "benchmark_experiment_board_row_key",
     "build_benchmark_candidate_source_boundary",
+    "build_benchmark_experiment_board",
     "build_benchmark_integrity_qualification",
     "build_native_codex_isolation_envelope",
     "build_native_goal_public_trajectory_summary",
@@ -131,16 +149,21 @@ __all__ = [
     "compact_native_codex_profile_receipt",
     "compact_native_goal_receipt",
     "compact_run_permission_policy_for_quota",
+    "default_benchmark_experiment_board_path",
     "filter_public_benchmark_artifact_paths",
     "inspect_native_codex_profile",
     "install_native_codex_profile",
     "materialize_public_benchmark_artifacts",
     "native_codex_app_server_environment",
     "native_codex_profile_environment",
+    "normalize_benchmark_experiment_board_row",
     "observe_native_goal_event",
+    "preview_benchmark_experiment_board_upsert",
     "probe_native_goal_process",
+    "read_benchmark_experiment_board_rows",
     "rebase_native_codex_loopx_workspace_state",
     "refresh_native_goal_status",
+    "render_benchmark_experiment_board_markdown",
     "render_native_codex_goal_prompt",
     "run_native_goal_process",
     "run_native_goal_process_until_terminal",
@@ -148,6 +171,7 @@ __all__ = [
     "run_native_goal_until_terminal",
     "select_exact_docker_container",
     "start_native_goal_turn",
+    "upsert_benchmark_experiment_board_row",
     "validate_run_permission_policy",
     "wait_native_goal_turn",
 ]

{loopx-0.4.9 → loopx-0.5.0}/loopx/capabilities/benchmark_toolkit/catalog_entry.py

@@ -24,6 +24,31 @@ BENCHMARK_TOOLKIT_CATALOG_ENTRY: dict[str, Any] = {
     ),
     "entry_command": "loopx benchmark --help",
     "commands": [
+        {
+            "command": (
+                "loopx benchmark experiment-board-show --goal-id <goal-id> "
+                "--format json"
+            ),
+            "purpose": (
+                "Read baseline, treatment, explore, countability, effort, and "
+                "insight status before selecting or launching another arm."
+            ),
+            "write_boundary": "read-only project-local public-safe domain state",
+        },
+        {
+            "command": (
+                "loopx benchmark experiment-board-upsert --goal-id <goal-id> "
+                "--row-json <compact-row.json> --execute --format json"
+            ),
+            "purpose": (
+                "Idempotently record one planned, running, or terminal compact run "
+                "row; omit --execute for a no-write preview."
+            ),
+            "write_boundary": (
+                "explicit project-local domain-state write; rejects raw paths, "
+                "logs, trajectories, hidden evaluation, credentials, and unknown fields"
+            ),
+        },
         {
             "command": (
                 "loopx benchmark integrity-qualification "
@@ -57,6 +82,37 @@ BENCHMARK_TOOLKIT_CATALOG_ENTRY: dict[str, Any] = {
             ),
         },
     ],
+    "agent_usage": {
+        "benchmark_start_hint": (
+            "At benchmark start, read the experiment board before selecting or "
+            "launching a case; update the same stable row at every material run "
+            "transition."
+        ),
+        "required_sequence": [
+            "read_experiment_board_before_launch_or_case_selection",
+            "upsert_preregistered_or_running_row_when_a_run_starts",
+            "upsert_terminal_score_countability_effort_and_insight_status",
+            "read_matched_comparisons_before_selecting_the_next_arm",
+        ],
+        "board_commands": {
+            "read": (
+                "loopx benchmark experiment-board-show --goal-id <goal-id> "
+                "--format json"
+            ),
+            "preview": (
+                "loopx benchmark experiment-board-upsert --goal-id <goal-id> "
+                "--row-json <compact-row.json> --format json"
+            ),
+            "write": (
+                "loopx benchmark experiment-board-upsert --goal-id <goal-id> "
+                "--row-json <compact-row.json> --execute --format json"
+            ),
+        },
+        "selection_rule": (
+            "Keep diagnostic-only explore rows separate and make paired claims "
+            "only from matched_pair_countable comparisons."
+        ),
+    },
     "post_run_case_analysis": {
         "benchmark_start_hint": (
             "When starting a benchmark, add one continuous_monitor todo that runs "
@@ -125,6 +181,16 @@ BENCHMARK_TOOLKIT_CATALOG_ENTRY: dict[str, Any] = {
         },
     },
     "implemented_protocols": [
+        {
+            "schema_version": "benchmark_experiment_board_row_v0",
+            "module": "loopx.capabilities.benchmark_toolkit.experiment_board",
+            "doc": "loopx/capabilities/benchmark_toolkit/README.md",
+        },
+        {
+            "schema_version": "benchmark_experiment_board_v0",
+            "module": "loopx.capabilities.benchmark_toolkit.experiment_board",
+            "doc": "loopx/capabilities/benchmark_toolkit/README.md",
+        },
         {
             "schema_version": "benchmark_integrity_qualification_v0",
             "module": "loopx.capabilities.benchmark_toolkit.integrity",
@@ -151,7 +217,12 @@ BENCHMARK_TOOLKIT_CATALOG_ENTRY: dict[str, Any] = {
             "doc": "loopx/capabilities/benchmark_toolkit/README.md",
         },
     ],
-    "smokes": ["python -m pytest tests/capabilities/test_benchmark_toolkit.py -q"],
+    "smokes": [
+        (
+            "python -m pytest tests/capabilities/test_benchmark_toolkit.py "
+            "tests/capabilities/test_benchmark_experiment_board.py -q"
+        )
+    ],
     "docs": ["loopx/capabilities/benchmark_toolkit/README.md"],
     "boundaries": [
         "The toolkit never grants runner, Docker, model, upload, submission, or publication authority; each effect remains separately gated.",
@@ -160,10 +231,10 @@ BENCHMARK_TOOLKIT_CATALOG_ENTRY: dict[str, Any] = {
         "Concurrent Docker runs must bind runtime evidence to one exact job-owned container; image-only discovery is not sufficient.",
         "Integrity qualification establishes countability eligibility only; an independent official result and matched experiment contract are still required.",
         "Post-run analyst access never widens the solving agent's evidence boundary or grants feedback reuse in another scored run.",
-        "Benchmark-family runners remain outside the active capability; the toolkit owns only provider-neutral permission, artifact, integrity, and analyst-brief policy.",
+        "The experiment board is a compact projection, not a score authority: benchmark-family runners and scoring adapters remain outside the active capability.",
     ],
     "next_real_step": (
-        "After each scored case, keep solver integrity separate and write one "
-        "private benchmark_case_insight_v0 from the complete evaluation evidence."
+        "Use the board before launch and after each scored case while keeping "
+        "solver integrity separate from private benchmark_case_insight_v0 analysis."
     ),
 }

{loopx-0.4.9 → loopx-0.5.0}/loopx/capabilities/benchmark_toolkit/environment_access.py

@@ -238,6 +238,7 @@ def _command_strings(function_name: str, raw_arguments: object) -> tuple[str, ..
 
 
 def _shell_segments(command: str) -> tuple[tuple[str, ...], ...]:
+    command = _without_shell_heredoc_bodies(command)
     try:
         lexer = shlex.shlex(command, posix=True, punctuation_chars=";&|\n")
         lexer.whitespace = " \t\r"
@@ -260,6 +261,67 @@ def _shell_segments(command: str) -> tuple[tuple[str, ...], ...]:
     return tuple(segments)
 
 
+def _heredoc_delimiters(command_line: str) -> tuple[tuple[str, bool], ...]:
+    """Return quoted-normalized heredoc delimiters declared by one shell line."""
+
+    try:
+        lexer = shlex.shlex(command_line, posix=True, punctuation_chars=";&|\n<>")
+        lexer.whitespace = " \t\r"
+        lexer.whitespace_split = True
+        lexer.commenters = ""
+        tokens = list(lexer)
+    except ValueError:
+        return ()
+
+    delimiters: list[tuple[str, bool]] = []
+    index = 0
+    while index < len(tokens):
+        if tokens[index] != "<<":
+            index += 1
+            continue
+        index += 1
+        strip_tabs = index < len(tokens) and tokens[index] == "-"
+        if strip_tabs:
+            index += 1
+        if index >= len(tokens):
+            break
+        delimiter = tokens[index]
+        if not strip_tabs and delimiter.startswith("-") and len(delimiter) > 1:
+            # ``shlex`` returns ``-EOF`` for the ordinary ``<<-EOF`` spelling.
+            strip_tabs = True
+            delimiter = delimiter[1:]
+        if delimiter and all(character not in ";&|\n<>" for character in delimiter):
+            delimiters.append((delimiter, strip_tabs))
+        index += 1
+    return tuple(delimiters)
+
+
+def _without_shell_heredoc_bodies(command: str) -> str:
+    """Exclude stdin payloads from shell-command classification.
+
+    Patch and file-generation tools commonly receive source text through a
+    heredoc.  That payload is data, not a sequence of shell commands, so a
+    source line such as ``env := value`` must not be classified as an executed
+    ``env`` credential probe.  Command lines before and after the payload stay
+    visible to the classifier.
+    """
+
+    visible_lines: list[str] = []
+    pending: list[tuple[str, bool]] = []
+    for raw_line in command.splitlines(keepends=True):
+        if pending:
+            delimiter, strip_tabs = pending[0]
+            candidate = raw_line.rstrip("\r\n")
+            if strip_tabs:
+                candidate = candidate.lstrip("\t")
+            if candidate == delimiter:
+                pending.pop(0)
+            continue
+        visible_lines.append(raw_line)
+        pending.extend(_heredoc_delimiters(raw_line))
+    return "".join(visible_lines)
+
+
 def _shell_executable(tokens: tuple[str, ...]) -> tuple[str, tuple[str, ...]]:
     remaining = list(tokens)
     while remaining and re.fullmatch(r"[A-Za-z_][A-Za-z0-9_]*=.*", remaining[0]):