feat: tool-call dedup plus shell/terminal hardening and editor fixes (#116)

* chore: clean completed tasks from todo.md

* fix: apply CodeRabbit auto-fixes

Fixed 1 file(s) based on 1 unresolved review comment.

Co-authored-by: CodeRabbit <noreply@coderabbit.ai>

* fix: harden shell, terminal, and plan-mode robustness

- Shell error briefs now surface the trailing output of a failed command
  (last non-empty lines rendered as plain text), so the collapsed worklog
  card explains why a command failed instead of only its exit code.
- Subagents no longer receive plan-mode workflow reminders; the reminder
  is root-only, since subagent toolsets usually exclude the plan-mode
  tools and the injection only invited calls to tools they don't have.
- The terminal cursor-position probe no longer risks hanging in raw mode
  on exit: reads are non-blocking during the probe and VMIN/VTIME are
  restored to canonical defaults so a cancelled probe can't wedge the tty.
- Reorder the welcome info block so session/storage details render last,
  and log the auto-save path failure instead of swallowing it silently.

* feat(soul): dedupe repeated tool calls within and across steps

Identical tool calls in one step now share the original task's result
instead of re-executing. Across steps, repeats are detected via
canonical JSON arguments (key order no longer defeats matching) and
nudged with sparse system-reminders at consecutive streaks of 3, 5,
and 8, reducing loop-thrash without hard-blocking legitimate retries.

Per-step state is armed inside the step-retry wrapper so a retried
step never awaits tasks cancelled by the failed attempt, and a D-Mail
revert clears the dedup seed since the reverted history no longer
contains those calls. Adds tool_call_dedup_detected telemetry plus a
dup_type property on tool_call events.

* chore: drop completed rename plan from todo, apply review nits

The pythinker-cli -> pythinker-code rename plan is fully realized
(root pyproject is pythinker-code, module is src/pythinker_code/, no
pythinker_cli references remain), so retire it from tasks/todo.md.

Review nits: annotate /reports with NoReturn and assert the injection
counter stays untouched in the subagent plan-mode suppression test.

* feat: editor bug fixes, Draft and auto save fixes.

* refactor: simplify dedup reminder path and tidy review nits

soul/toolset: append the cross-step dedup reminder inline at the tool task's return instead of behind a second wrapper task. Behaviour is unchanged (toolset tests pass) but the returned task is now the tool task itself, with one fewer task and closure per reminder. Drop the now-unused default on _append_reminder_to_return_value and annotate begin_step/end_step with the existing ToolCallKey alias.

tools/utils: tail() docstring no longer over-specifies the buffer as stderr (it holds mixed command output); rename stripped -> rstripped to reflect that it holds an rstrip() result.

* fix: address CodeRabbit review on dedup reminder and test

Bound the canonical arguments echoed in the strong dedup reminder to a 256-char preview so large-payload tools (WriteFile, MultiEdit) don't re-inject their whole body into context on every repeat; exact identity is still carried by the args_hash dedup telemetry.

Rewrite test_begin_end_step to assert observable behaviour (handle()/end_step()/dedup_triggered) instead of poking private _current_step_* internals.

* fix(test): narrow handle() result to Task before await (pyright)

handle() returns Task[ToolResult] | ToolResult; assert isinstance Task before awaiting, matching the other dedup tests, so pyright's check job passes.

---------

Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
Co-authored-by: CodeRabbit <noreply@coderabbit.ai>

Author: 3 people

CHANGELOG.md

@@ -15,6 +15,9 @@ GitHub Releases page; `0.8.0` is the new starting line.
 
 ## Unreleased
 
+- **Shell error briefs now show the trailing output of a failed command.** When a `Shell`/`Terminal` command exits non-zero, times out, or is killed by a signal, the collapsed worklog card appended only `Failed with exit code: N`; you had to expand the result to see *why*. The brief now includes the last few non-empty output lines (e.g. the stderr message), rendered as plain text so shell metacharacters (backticks, `#`, `*`) and line breaks are preserved verbatim instead of being reflowed as Markdown.
+- **Subagents no longer receive plan-mode workflow reminders.** Plan mode is a session-wide flag shared with subagents (so it persists across resume), but subagent toolsets usually exclude `EnterPlanMode`/`ExitPlanMode`. Injecting the plan-mode reminder into a subagent only invited hallucinated calls to tools it doesn't have; the reminder is now root-only.
+- **Terminal no longer risks hanging in raw mode on exit.** The cursor-position probe left `stdin` in cbreak mode and could block in an uninterruptible `os.read()` if cancelled mid-probe (e.g. a race with prompt_toolkit's reader on shutdown). Reads are now non-blocking during the probe and `VMIN`/`VTIME` are restored to canonical defaults, so a hang or crash can't leave the terminal wedged.
 - **New `/goal` command: goal-driven execution ported from Codex CLI.** `/goal <objective>` sets a persistent thread goal the agent pursues across turns until it is verifiably complete. The objective is stored in session state (survives restarts and context compaction), kicks off work immediately with a success-criteria derivation prompt, and is re-injected on later turns as a continuation reminder carrying Codex's fidelity rules (no scope-shrinking, no easier-to-test substitutes) and evidence-based completion audit — the agent may only claim completion after proving every requirement against current state, and the user confirms with `/goal clear`. Subcommands: `view`, `pause`, `resume`, `clear`. Objectives are injected as untrusted data (`<objective>` framing), never as higher-priority instructions.
 - **New `/best-practices` command (alias `/bp`).** Injects opt-in engineering best-practice guidance distilled from the Codex CLI system prompts — code-change discipline, dirty-worktree safety (never revert changes you didn't make), specific-to-broad testing strategy, todo hygiene, progress-update cadence, debugging methodology, and final-answer style — into the session context without consuming a turn. `/best-practices <section>` injects a single section.
 - **SetTodoList nudges the single-`in_progress` discipline.** Todo lists with more than one `in_progress` item now get a corrective notice (ported from Codex's plan-tool contract, softened because parallel-subagent fan-out legitimately tracks one `in_progress` sub-todo per running child), and the system prompt gains matching status-discipline guidance: no single-step lists, no `pending`→`done` jumps, no batch-completing after the fact.

docs/en/reference/pythinker-vis.md

@@ -18,7 +18,7 @@ The server automatically opens a browser after startup. The default address is `
 
 If the default port is in use, the server will pick the next available port (by default `5495`–`5504`) and print the access URL in the terminal.
 
-You can also type `/vis` in the interactive shell to switch directly from the current session to the Visualizer.
+You can also type `/reports` in the interactive shell to switch directly from the current session to the Visualizer.
 
 ## Command-line options
 

docs/en/reference/slash-commands.md

@@ -346,9 +346,9 @@ Auto mode skips all approval confirmations and removes the clarifying-question s
 
 Switch to Web UI. Pythinker Code will start a Web UI server and open the current session in your browser, allowing you to continue the conversation in the Web UI. See [Web UI](./pythinker-web.md) for details.
 
-### `/vis`
+### `/reports`
 
-Switch to the Agent Tracing Visualizer. Pythinker Code will start the visualizer server and open the current session's tracing view in the browser, where you can inspect Wire event timelines, context messages, and usage statistics. See [Agent Tracing Visualizer](./pythinker-vis.md) for details.
+Open session reports in the Agent Tracing Visualizer. Pythinker Code will start the visualizer server and open the current session's tracing view in the browser, where you can inspect Wire event timelines, context messages, and usage statistics. See [Agent Tracing Visualizer](./pythinker-vis.md) for details.
 
 ## Command completion
 

src/pythinker_code/acp/tools.py

@@ -149,20 +149,22 @@ async def __call__(self, params: ShellParams) -> ToolReturnValue:
                 else ""
             )
 
+            tail = builder.tail()
+            tail_suffix = f"\n{tail}" if tail else ""
             if timed_out:
                 return builder.error(
                     f"Command killed by timeout ({timeout_label}){truncated_note}",
-                    brief=f"Killed by timeout ({timeout_label})",
+                    brief=f"Killed by timeout ({timeout_label}){tail_suffix}",
                 )
             if exit_signal:
                 return builder.error(
                     f"Command terminated by signal: {exit_signal}.{truncated_note}",
-                    brief=f"Signal: {exit_signal}",
+                    brief=f"Signal: {exit_signal}{tail_suffix}",
                 )
             if exit_code not in (None, 0):
                 return builder.error(
                     f"Command failed with exit code: {exit_code}.{truncated_note}",
-                    brief=f"Failed with exit code: {exit_code}",
+                    brief=f"Failed with exit code: {exit_code}{tail_suffix}",
                 )
             return builder.ok(f"Command executed successfully.{truncated_note}")
         finally:

src/pythinker_code/app.py

@@ -789,7 +789,6 @@ async def run_shell(
         branch_name = _safe_git_branch(work_dir)
         if branch_name:
             welcome_info.append(WelcomeInfoItem(name="Branch", value=branch_name))
-        welcome_info.append(WelcomeInfoItem(name="Session", value=self._runtime.session.id))
         if notice := _resumed_unsupervised_notice(
             resumed=self._runtime.resumed,
             yolo=self._runtime.approval.is_yolo(),
@@ -798,14 +797,6 @@ async def run_shell(
             welcome_info.append(
                 WelcomeInfoItem(name="Mode", value=notice, level=WelcomeInfoItem.Level.WARN)
             )
-        try:
-            auto_save_path = str(
-                shorten_home(HostPath.unsafe_from_local_path(self._runtime.session.context_file))
-            )
-        except Exception:
-            auto_save_path = ""
-        if auto_save_path:
-            welcome_info.append(WelcomeInfoItem(name="Auto-save", value=auto_save_path))
         if base_url := self._env_overrides.get("PYTHINKER_BASE_URL"):
             welcome_info.append(
                 WelcomeInfoItem(
@@ -861,6 +852,18 @@ async def run_shell(
                         level=WelcomeInfoItem.Level.WARN,
                     )
                 )
+        # Session persistence details come last — workspace and model identity
+        # read first, storage internals stay at the bottom of the facts block.
+        welcome_info.append(WelcomeInfoItem(name="Session", value=self._runtime.session.id))
+        try:
+            auto_save_path = str(
+                shorten_home(HostPath.unsafe_from_local_path(self._runtime.session.context_file))
+            )
+        except Exception:
+            logger.debug("Failed to compute auto-save display path", exc_info=True)
+            auto_save_path = ""
+        if auto_save_path:
+            welcome_info.append(WelcomeInfoItem(name="Auto-save", value=auto_save_path))
         welcome_info.append(
             WelcomeInfoItem(
                 name="Tip",

src/pythinker_code/auth/openai/models.py

@@ -175,7 +175,7 @@ def _parse_chatgpt_models_payload(payload: object) -> list[ModelInfo]:
     raw_models = payload_object.get("models")
     if not isinstance(raw_models, list):
         # Keep a small compatibility path in case OpenAI ever aligns this with
-        # the public /v1/models shape. ChatGPT Codex currently returns
+        # the public /v1/models shape. The ChatGPT endpoint currently returns
         # {"models": [{"slug": ...}]}.
         raw_models = payload_object.get("data")
     if not isinstance(raw_models, list):

src/pythinker_code/auth/openai/oauth_client.py

@@ -184,7 +184,7 @@ def _token_from_openai_response(payload: dict[str, Any]) -> OAuthToken:
     # it lives inside the OAuth JWT claims under
     # `https://api.openai.com/auth.chatgpt_account_id`. Hoist it onto the
     # response so OAuthToken.from_response() picks it up. Without this the
-    # ChatGPT usage adapter, model catalog endpoint, and Codex request headers
+    # ChatGPT usage adapter, model catalog endpoint, and request headers
     # cannot scope requests to the active Plus/Pro account.
     if "account_id" not in normalized:
         jwt_token = payload.get("id_token") or payload.get("access_token")

src/pythinker_code/auth/platforms.py

@@ -176,7 +176,7 @@ def _select_retry_api_keys(
 
 
 def _openai_fallback_models(platform_id: str) -> list[ModelInfo] | None:
-    # ChatGPT Codex model availability is subscription/account-specific. Do not
+    # ChatGPT model availability is subscription/account-specific. Do not
     # replace the user's live catalog with a static fallback; stale fallback
     # slugs surface as 400 "model is not supported with a ChatGPT account".
     if platform_id == OPENAI_CHATGPT_PLATFORM_ID:

src/pythinker_code/cli/__init__.py

@@ -911,7 +911,7 @@ async def _run(session_id: str | None, prefill_text: str | None = None) -> tuple
             scratchpad_status = await ensure_git_excluded(work_dir)
 
             # Sweep accumulated state on startup (best-effort, non-blocking).
-            # Mirrors Claude Code's cleanupPeriodDays=30 model.
+            # Default 30-day retention sweep.
             _retention = config.session_retention_days if isinstance(config, Config) else 30
             await asyncio.to_thread(sweep_old_sessions, _retention)
             await asyncio.to_thread(sweep_old_plans, _retention)

src/pythinker_code/hooks/runner.py

@@ -110,7 +110,7 @@ async def run_hook(
 
 
 def _extract_additional_context(parsed: dict[str, Any], hook_output: dict[str, Any]) -> str:
-    """Extract Claude-Code-style additionalContext from JSON hook output."""
+    """Extract ``additionalContext`` from JSON hook output."""
     candidates = (hook_output.get("additionalContext"), parsed.get("additionalContext"))
     for value in candidates:
         if isinstance(value, str) and value.strip():