noah-code 0.5.1__tar.gz → 0.7.0__tar.gz

{noah_code-0.5.1 → noah_code-0.7.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: noah-code
-Version: 0.5.1
+Version: 0.7.0
 Summary: Noah Code terminal coding agent, built on NVIDIA OO Agents (NOOA)
 Project-URL: Homepage, https://github.com/skundu42/noah-code
 Project-URL: Documentation, https://github.com/skundu42/noah-code#readme
@@ -28,6 +28,7 @@ Requires-Dist: pydantic<3.0.0,>=2.5.0
 Requires-Dist: pyyaml<7.0.0,>=6.0.0
 Requires-Dist: rich<16.0.0,>=13.0.0
 Requires-Dist: textual<9.0.0,>=1.0.0
+Requires-Dist: tomlkit<1.0.0,>=0.13.2
 Provides-Extra: dev
 Requires-Dist: mypy>=2.3.1; extra == 'dev'
 Requires-Dist: pytest-asyncio>=0.24.0; extra == 'dev'
@@ -61,8 +62,8 @@ Description-Content-Type: text/markdown
 
 <p align="center"><sub>The real Textual interface, captured from a deterministic Noah Code session.</sub></p>
 
-Noah keeps the conversation central while the context rail tracks live work, Git changes, session,
-model usage, and the current plan. Tool execution stays visible, completed work compacts into
+Noah keeps the conversation central while the sidebar tracks the current action, changed files,
+and the active plan. Tool execution stays visible, completed work compacts into
 readable records, and every session remains scoped to its repository.
 
 Built on the [NVIDIA OO Agents (NOOA)](https://github.com/NVIDIA-NeMo/labs-OO-Agents) runtime.
@@ -136,6 +137,9 @@ noah --mode plan .
 # Run one task and exit
 noah run "Explain how authentication is wired" .
 
+# Emit one structured result on stdout (progress goes to stderr)
+noah run --json "Fix the failing unit test" .
+
 # Allow actions that would normally ask; explicit deny rules still apply
 noah run --auto "Fix the failing unit test" .
 
@@ -147,6 +151,13 @@ noah --session SESSION_ID .
 noah --console .
 ```
 
+`run --json` reports `status` (`completed`, `needs_input`, `failed`, or `cancelled`),
+`exit_code`, `explanation`, `session_id`, `run_id`, session `usage`, and recorded `checks`.
+Check `status` when automating: a request for input keeps the existing exit code 0,
+but its status is `needs_input`. Check results are separate from task completion;
+an empty list means no checks were recorded. Errors keep nonzero exit codes and
+Ctrl+C returns 130. Delegated agents also retain `needs_input` in the work ledger.
+
 Check the installation and resolved configuration with:
 
 ```bash
@@ -166,20 +177,33 @@ Type `/` to search the full command and configuration reference. The most common
 | Control | Action |
 | --- | --- |
 | `Enter` | Send the current prompt or accept a selected suggestion |
-| `Shift+Enter` | Insert a newline |
+| `Shift+Enter` / `Ctrl+J` | Insert a newline |
+| `Alt+Enter` | Expand or collapse the composer |
+| `Alt+Z` | Restore a displaced draft |
+| Click / arrows in the prompt | Position the editing cursor |
+| `Ctrl+A` / `Ctrl+X` / `Ctrl+V` | Select all / cut / paste prompt text |
+| `Ctrl+Z` / `Ctrl+Y` | Undo / redo prompt edits |
+| `Ctrl+Home` / `Ctrl+End` | Jump to the beginning / end of the prompt; Shift selects |
 | Drag, then `Cmd+C` / `Ctrl+Shift+C` | Select and copy TUI text |
 | `Ctrl+Shift+C` with no selection | Copy the latest Noah reply |
 | `Ctrl+G` | Open the searchable skills picker |
+| `Ctrl+P` | Search and run slash commands |
 | `Ctrl+L` | Open the model picker |
 | `Ctrl+R` | Search and recall a prior prompt |
 | `Ctrl+T` | Expand or collapse live tool output |
-| `Shift+Tab` | Open the reasoning-effort picker |
+| `Alt+E` | Open the reasoning-effort picker |
 | `Alt+Up` | Recall the newest queued prompt for editing |
 | `Ctrl+]` | Return to live transcript output |
-| `Tab` | Switch between build and plan mode |
+| `Tab` / `Shift+Tab` | Navigate focus; Tab completes suggestions |
+| `Enter` or click a command | Run the selected slash command or open its picker; required arguments stay editable |
+| `Ctrl+B` | Switch between build and plan mode |
+| `Ctrl+D` | Review changes while working or idle |
+| `Ctrl+C` | Stop the current run and pause queued follow-ups |
 | `F2` | Open execution activity |
 | `F3` | Open paginated conversation history |
 | `F4` | Open the live agent, terminal, and job ledger |
+| `F5` | Edit, reorder, resume, or discard queued follow-ups |
+| `F8` | Show or hide the sidebar |
 | `Shift+F7` | Focus or leave the scrollable context rail |
 | `/model` | Configure a provider or switch the session model |
 | `/theme` | Choose Atom One Dark, Noah Ocean, Graphite, or High Contrast |
@@ -247,7 +271,7 @@ are configured under `[reliability.retries]`.
 ## Updates
 
 Noah checks PyPI for new versions at most once every 24 hours. New TUI sessions show a temporary
-banner when an update is available and keep the version visible in the context rail. Installation
+banner when an update is available, with `F6` opening its details. Installation
 remains explicit by default:
 
 ```bash

{noah_code-0.5.1 → noah_code-0.7.0}/README.md

@@ -16,8 +16,8 @@
 
 <p align="center"><sub>The real Textual interface, captured from a deterministic Noah Code session.</sub></p>
 
-Noah keeps the conversation central while the context rail tracks live work, Git changes, session,
-model usage, and the current plan. Tool execution stays visible, completed work compacts into
+Noah keeps the conversation central while the sidebar tracks the current action, changed files,
+and the active plan. Tool execution stays visible, completed work compacts into
 readable records, and every session remains scoped to its repository.
 
 Built on the [NVIDIA OO Agents (NOOA)](https://github.com/NVIDIA-NeMo/labs-OO-Agents) runtime.
@@ -91,6 +91,9 @@ noah --mode plan .
 # Run one task and exit
 noah run "Explain how authentication is wired" .
 
+# Emit one structured result on stdout (progress goes to stderr)
+noah run --json "Fix the failing unit test" .
+
 # Allow actions that would normally ask; explicit deny rules still apply
 noah run --auto "Fix the failing unit test" .
 
@@ -102,6 +105,13 @@ noah --session SESSION_ID .
 noah --console .
 ```
 
+`run --json` reports `status` (`completed`, `needs_input`, `failed`, or `cancelled`),
+`exit_code`, `explanation`, `session_id`, `run_id`, session `usage`, and recorded `checks`.
+Check `status` when automating: a request for input keeps the existing exit code 0,
+but its status is `needs_input`. Check results are separate from task completion;
+an empty list means no checks were recorded. Errors keep nonzero exit codes and
+Ctrl+C returns 130. Delegated agents also retain `needs_input` in the work ledger.
+
 Check the installation and resolved configuration with:
 
 ```bash
@@ -121,20 +131,33 @@ Type `/` to search the full command and configuration reference. The most common
 | Control | Action |
 | --- | --- |
 | `Enter` | Send the current prompt or accept a selected suggestion |
-| `Shift+Enter` | Insert a newline |
+| `Shift+Enter` / `Ctrl+J` | Insert a newline |
+| `Alt+Enter` | Expand or collapse the composer |
+| `Alt+Z` | Restore a displaced draft |
+| Click / arrows in the prompt | Position the editing cursor |
+| `Ctrl+A` / `Ctrl+X` / `Ctrl+V` | Select all / cut / paste prompt text |
+| `Ctrl+Z` / `Ctrl+Y` | Undo / redo prompt edits |
+| `Ctrl+Home` / `Ctrl+End` | Jump to the beginning / end of the prompt; Shift selects |
 | Drag, then `Cmd+C` / `Ctrl+Shift+C` | Select and copy TUI text |
 | `Ctrl+Shift+C` with no selection | Copy the latest Noah reply |
 | `Ctrl+G` | Open the searchable skills picker |
+| `Ctrl+P` | Search and run slash commands |
 | `Ctrl+L` | Open the model picker |
 | `Ctrl+R` | Search and recall a prior prompt |
 | `Ctrl+T` | Expand or collapse live tool output |
-| `Shift+Tab` | Open the reasoning-effort picker |
+| `Alt+E` | Open the reasoning-effort picker |
 | `Alt+Up` | Recall the newest queued prompt for editing |
 | `Ctrl+]` | Return to live transcript output |
-| `Tab` | Switch between build and plan mode |
+| `Tab` / `Shift+Tab` | Navigate focus; Tab completes suggestions |
+| `Enter` or click a command | Run the selected slash command or open its picker; required arguments stay editable |
+| `Ctrl+B` | Switch between build and plan mode |
+| `Ctrl+D` | Review changes while working or idle |
+| `Ctrl+C` | Stop the current run and pause queued follow-ups |
 | `F2` | Open execution activity |
 | `F3` | Open paginated conversation history |
 | `F4` | Open the live agent, terminal, and job ledger |
+| `F5` | Edit, reorder, resume, or discard queued follow-ups |
+| `F8` | Show or hide the sidebar |
 | `Shift+F7` | Focus or leave the scrollable context rail |
 | `/model` | Configure a provider or switch the session model |
 | `/theme` | Choose Atom One Dark, Noah Ocean, Graphite, or High Contrast |
@@ -202,7 +225,7 @@ are configured under `[reliability.retries]`.
 ## Updates
 
 Noah checks PyPI for new versions at most once every 24 hours. New TUI sessions show a temporary
-banner when an update is available and keep the version visible in the context rail. Installation
+banner when an update is available, with `F6` opening its details. Installation
 remains explicit by default:
 
 ```bash

{noah_code-0.5.1 → noah_code-0.7.0}/docs/configuration.md

@@ -219,12 +219,14 @@ supported by the OTLP exporters; keep authentication headers in the environment
 secret store rather than TOML.
 
 Repository-controlled configuration cannot weaken the host trust boundary. Project config is
-ignored for `auto_approve`, `budget`, `efficiency`, `enabled_skills`, `hooks`, `lsp`, `mcp`,
+ignored for `auto_approve`, `yolo`, `budget`, `checkpoints`, `efficiency`, `enabled_skills`, `hooks`, `lsp`, `mcp`,
 `permission_rules`, `processes`, `reliability`, `session_dir`, `tracing`, `updates`, and
 `unsafe_inprocess_code_execution`. Put those settings in trusted user config, the environment, or
 an explicit CLI flag. Language-server overrides and hooks are user-only because they launch local
-executables; reliability and budget settings are user-only so repository content cannot weaken
-host limits.
+executables; reliability, budget, and checkpoint settings are user-only so repository content cannot
+weaken host limits or recovery protections. Project config may select `mode = "plan"`, but choosing
+build mode requires trusted configuration or an explicit CLI flag. Newly added settings are user-only
+unless explicitly approved for project configuration.
 
 A user-configured `permission_rules` array replaces the default rule array. Copy forward every
 default you still want before adding overrides. Hard secret, destructive-shell, and plan-mode
@@ -362,8 +364,8 @@ Permission rules are evaluated in order, and the last matching rule wins. The de
 - Keeps file tools inside the active workspace and asks before MCP access.
 - Denies plan-mode mutations regardless of broader allow rules. Plan mode may still run
   read-only subagents. Its shell allowlist accepts only literal, unqualified read-only programs;
-  test collection and interpreter execution are not treated as read-only because they can load
-  repository code.
+  arguments are checked for write-capable options. Test collection and interpreter execution,
+  including arbitrary awk programs, are not treated as read-only because they can execute code.
 
 `--auto` changes routine ask decisions to allow but never overrides an explicit deny.
 Elevated-risk commands such as file removal, downloads, and package installation still require
@@ -384,7 +386,7 @@ sh install.sh
 ```
 
 Installs created by the one-line command check PyPI at most once every 24 hours. If a release is
-available, a new TUI session shows a temporary banner and keeps the version in the context rail.
+available, a new TUI session shows a temporary banner with `F6` opening its details.
 Run `noah update` when ready. When an update is installed, Noah exits before starting the task so
 it cannot mix old and new runtime modules. Rerun the command to continue on the new version.
 

{noah_code-0.5.1 → noah_code-0.7.0}/docs/interactive-reference.md

@@ -4,60 +4,78 @@
 
 | Key | Action |
 |-----|--------|
-| `Enter` | Send the current message, or queue a follow-up while a turn is running |
-| `Shift+Enter` | Insert a newline without sending |
-| `Tab` | Toggle `build`/`plan` mode; accept the highlighted slash option while suggestions are open |
-| `Ctrl+P` | Open the command palette |
+| `Enter` | Run the highlighted command, send the current message, or queue a follow-up while a turn is running |
+| `Shift+Enter` or `Ctrl+J` | Insert a newline without sending |
+| `Alt+Enter` | Expand or collapse the composer |
+| `Alt+Z` | Restore the draft saved before a palette or history selection |
+| `Tab` / `Shift+Tab` | Move focus forward/back; Tab completes an open suggestion |
+| `Ctrl+B` | Switch between build and plan mode |
+| `Ctrl+P` | Search and run slash commands |
 | `Ctrl+G` | Open the searchable skills picker |
 | `Ctrl+L` | Open the model picker |
 | `Ctrl+O` | Open the session picker |
 | `Ctrl+R` | Search prior prompts and recall one into the composer |
 | `Ctrl+T` | Expand or collapse live tool output |
 | `Alt+Up` | Recall the newest queued prompt into the composer |
-| `Shift+Tab` | Open the reasoning-effort picker |
+| `Alt+E` | Open the reasoning-effort picker |
+| `Ctrl+D` | Review changed files, including while work is running |
 | `Ctrl+N` | Start a new session |
-| `Ctrl+C` | Cancel the active turn and clear queued follow-ups; press twice while idle to quit |
+| `Ctrl+C` | Stop the active turn and pause queued follow-ups, keeping attachments; press twice while idle to quit |
 | `Ctrl+Q` | Quit |
 | `F1` or `?` | Show help |
-| `F2` | Open recent activity and full captured output |
-| `F3` | Open paginated persisted conversation history |
+| `F2` | Search recent activity and inspect captured output |
+| `F3` | Search loaded conversation history; Ctrl+Home loads older messages |
 | `F4` | Open the live work ledger for agents, terminals, and background jobs |
+| `F5` | Manage queued prompts: E edit, R resume, X discard all |
+| `F6` | Inspect the latest notice or error |
+| `F7` | Inspect context sources |
+| `F8` | Show or hide the sidebar |
 | `Shift+F7` | Focus or leave the context rail on wide terminals |
 | `Ctrl+]` | Return to live transcript output and clear the new-output counter |
 
 At an approval prompt, press `1` to approve once, `2` to remember the approval for the current
-session, or `3`/`Esc` to reject it.
+session, or `3`/`Esc` to reject it. The dialog explains the matching scope and keeps the
+full target scrollable. Question cards accept numbered choices and show question progress;
+`Esc` skips that question.
 
 The TUI uses Atom One Dark by default and also includes Noah Ocean, Graphite, and High Contrast.
-Switch and persist the active palette with `/theme`. Its conversation-first layout adapts to
-terminal size. At 110 columns or wider, a change-ledger rail prioritizes the current action, Git
-branch and staged/modified/new counts, session, model usage, and up to six open or blocked todos.
-Narrow terminals retain the transcript, live activity, suggestions, and composer without the rail.
-Terminals 25 rows high or shorter use compact spacing.
-
-The line below the composer keeps the active keyboard guidance on the left and, when space permits,
+Switch and persist the active palette with `/theme`; dialogs and tool output use the same
+semantic colors. The sidebar appears automatically at 128 columns and 24 rows or larger,
+prioritizing the current action, changed filenames, and the active plan. Empty work sections
+stay hidden. `F8` toggles the sidebar and remembers that choice for the app run; below 100
+columns it stays hidden. Narrow terminals keep the full conversation width. Terminals 25
+rows high or shorter use compact spacing. Model and mode remain in the header; `/status`,
+`/tokens`, and the existing inspectors provide full metadata.
+
+The line below the composer keeps clickable, focused-pane keyboard guidance on the left and, when space permits,
 session input/output tokens, cache hit rate, and estimated cost on the right. This telemetry remains
 available even when the wide context rail is hidden.
 
 Live tool output stays in a compact two-line drawer so long commands do not push the conversation
-away. Press `Ctrl+T` to expand the drawer in place; `F2` retains the complete output after the tool
-finishes. The drawer replaces the animated working banner while a visible tool is active, avoiding
+away. Press `Ctrl+T` to expand the drawer in place; `F2` retains bounded captured output after the tool
+finishes, explicitly marking omitted middle output when the capture limit is reached. The drawer replaces the animated working banner while a visible tool is active, avoiding
 duplicate status lines; the banner returns between tools.
 
 Type `/` in the composer to open the inline command list; the list remains visible and filters
-continuously. Use `Up`/`Down` to highlight a command, `Enter` or `Tab` to complete it, and `Esc` to
-close the list. Press `Enter` again to run the completed command. Typing `/config` expands the list
-to every resolved configuration path and its current redacted value.
+continuously. Use `Up`/`Down` to highlight a command, then press `Enter` or click once to run it
+or open its picker. `Tab` completes the command for editing, and `Esc` closes the list. Commands
+requiring arguments and file mentions remain editable. Typing `/config` expands the list to every
+resolved configuration path and its current redacted value. `Ctrl+P` offers the same single-action
+selection in a searchable command picker.
 
 Press `Ctrl+R` to search prompts already shown in the current session. Choosing one loads it into
-the composer without sending it; pressing `Esc` keeps the current draft unchanged.
-
-Until a session has its first user prompt, the main pane keeps the large Noah wordmark centered. Startup,
-repository changes, model, mode, usage, and update state live in the context rail on wide
-terminals. The rail scrolls when its sections exceed the available rows; press `Shift+F7`, then use
-arrows, Page Up/Down, Home, or End without leaving the keyboard. Git status is collected in a
-background worker at startup and turn boundaries; the animated Noah path updates only the working
-banner and live activity. Existing sessions with user history restore their transcript normally.
+the composer without sending it and saves the displaced draft for `Alt+Z`. Palette choices
+use the same draft protection; skill selections insert at the cursor. Pressing `Esc` leaves
+the current draft unchanged. Commands blocked while busy remain editable.
+
+Until a session has its first user prompt, the main pane keeps the Noah wordmark centered.
+The sidebar scrolls when needed; press Tab to focus it or use `Shift+F7`, then arrows,
+Page Up/Down, Home, or End. Git status is collected in a background worker at startup and
+turn boundaries, and every five seconds while working with the sidebar visible. Existing
+sessions restore their recent transcript. F3 preserves reading position when older pages
+load; its search covers the messages loaded so far. F2 searches captured action labels,
+commands, results, and output. Failed tool attempts remain visible in the conversation
+with an F2 details hint.
 
 Drag across transcript, activity, diff, or history text to select it. `Cmd+C` on macOS or
 `Ctrl+Shift+C` in other terminals copies the selection; when there is no selection, the same
@@ -72,6 +90,28 @@ Inside the composer, `Cmd+C`/`Ctrl+C` copies the composer's
 own selection; `Ctrl+C` with no composer selection keeps its cancel-turn behavior. Selection uses a
 high-contrast highlight in every Noah theme.
 
+The prompt input supports mouse placement, drag selection, and normal text editing:
+
+| Prompt control | Action |
+| --- | --- |
+| Click / arrow keys | Place or move the cursor within the prompt |
+| `Shift` + arrow keys | Extend a selection |
+| `Ctrl+A` / `Cmd+A` | Select the entire prompt |
+| `Ctrl+X` / `Cmd+X` | Cut selected text; with no selection, cut the current line |
+| `Ctrl+C` / `Cmd+C` | Copy selected prompt text |
+| `Ctrl+V` / `Cmd+V` | Paste, replacing the selected text |
+| `Ctrl+Z` / `Ctrl+Y` | Undo / redo prompt edits |
+| `Home` / `End` | Move to the beginning / end of the current line |
+| `Ctrl+Home` / `Ctrl+End` | Move to the beginning / end of the entire prompt |
+| `Ctrl+Shift+Home` / `Ctrl+Shift+End` | Select to the beginning / end of the entire prompt |
+| `Ctrl+Left` / `Ctrl+Right` | Move by word; hold Shift to select |
+| `Alt+Enter` | Expand the prompt for longer edits |
+
+`Ctrl+A` now selects all; use `Home` for the former line-start behavior. Prompt undo is
+separate from `/undo`, which reverts workspace edits. A paste waits for an in-progress
+cut/copy to reach the clipboard; if the draft, selection, focus, or session changes while
+the clipboard is being read, that paste is cancelled to protect the newer input.
+
 ### Mid-turn follow-ups
 
 While Noah is working, the composer stays open. `Enter` queues the current text instead of starting
@@ -85,15 +125,18 @@ The queue holds at most 100 items. A 101st `Enter` drops the oldest and status-p
 `steer dropped oldest`. `@path` mentions and `/attach` paths expand when the item is injected, not
 when it is queued. A follow-up that names files Noah cannot resolve is dropped; later items stay.
 Sequenced items are persisted in the session runtime database, so an unexpected process exit does
-not lose them. `Ctrl+C` cancels the turn and clears the queue. Switching or starting a session also
-clears it.
+not lose them. `Ctrl+C` stops the current turn and pauses delivery, keeping queued prompts
+and pending attachments. Pause state, order, and attachments survive reopening the session.
+F5 lets you edit a selected prompt, resume delivery with R, or explicitly discard everything
+with X. Switching or starting a session clears the active queue; durable state belongs to
+its originating session.
 
 Approval and `ask.question` modals keep the composer. Queueing resumes after the modal closes.
 
 These slash commands still run while a turn is in progress: `/status`, `/tokens`, `/todos`,
-`/health`, `/help`, `/trace`, `/work`, and `/terminals`. `/attach PATH` remembers the file for the next queued follow-up.
-`/exit` cancels the turn (and the queue) then leaves. Mutating commands wait until the turn
-finishes, including `/undo`, `/redo`, `/mode`, `/model`, `/diff`, `/new`, `/sessions`, `/worktree`,
+`/health`, `/help`, `/trace`, `/work`, `/diff`, and `/terminals`. `/attach PATH` remembers the file for the next queued follow-up.
+`/exit` stops the turn and leaves with queued input preserved. Mutating commands wait until the turn
+finishes, including `/undo`, `/redo`, `/mode`, `/model`, `/new`, `/sessions`, `/worktree`,
 `/pr`, `/plan`, `/memory`, and `/compact`.
 
 Tool and shell output is batched into a live execution panel instead of forcing one full-screen
@@ -104,6 +147,23 @@ Consecutive reads or writes merge into a single line (`✓ Read a.py, b.py +1`)
 compact. `F2` retains the latest 100 activity records, bounded by the configured
 `max_output_chars` per activity.
 
+
+### Reviewing results
+
+`Ctrl+D` or `/diff` opens the changed-file list first. Patches and editor diagnostics load
+when selected, with separate capture timestamps; Ctrl+R refreshes the list. Review is
+available during a run, with revert and undo disabled until work is idle. Type `/` to
+filter filenames, J/K to change files, N/P to move between hunks, V to toggle editor
+diagnostics, and O to open the selected
+file using `$VISUAL` or `$EDITOR` (falling back to `vi`). Large patch previews explicitly
+state when truncated. Editor diagnostics describe the current worktree, including when
+viewing staged changes.
+
+Completion receipts show recorded check commands and their actual exit statuses, separate
+from editor diagnostics. They include observable changed-file counts and advertise journal
+undo only when its preflight succeeds. When results are unavailable, the receipt says so;
+compound shell expressions are not treated as individual passing test commands.
+
 ## Built-in slash commands
 
 | Command | Purpose |

noah_code-0.7.0/docs/releases/v0.6.0.md

@@ -0,0 +1,68 @@
+# Noah Code v0.6.0
+
+Noah now preserves drafts and queued work, makes changes easier to review while the agent runs,
+and reports recorded check results alongside accurate undo availability. This release also fixes
+permission boundaries, crash recovery, checkpoint restoration, and resumed-session continuity.
+
+## Input and navigation
+
+- Palette and history selections save the displaced draft for `Alt+Z`; skills insert at the
+  cursor. Blocked commands and failed queue writes leave the prompt editable.
+- `Ctrl+C` stops the current run and pauses queued prompts while retaining attachments.
+  Paused state, delivery order, and attachments survive reopening. `F5` edits a selected prompt,
+  resumes delivery, or explicitly discards waiting input.
+- `Tab` and `Shift+Tab` move focus; Tab still completes open suggestions. Use `Ctrl+B` for
+  Build/Plan, `Alt+E` for reasoning, `Alt+Enter` to expand the composer, and `Ctrl+J` as a newline
+  alternative to Shift+Enter. These replace the previous Tab and Shift+Tab mode/settings shortcuts.
+- The sidebar prioritizes current work, changed filenames, and the active plan. It appears
+  automatically at 128 columns and 24 rows or larger; `F8` toggles it. Footer actions follow focus.
+- `F2` searches captured activity; `F3` searches loaded conversation history and preserves reading
+  position when loading older pages. Failed tool attempts remain visible with a details hint.
+- Approval details scroll without hiding the decision buttons. Question cards support numbered
+  choices and progress. Dialogs use the selected theme's semantic colors.
+
+## Review and check results
+
+- `Ctrl+D` or `/diff` opens the file list immediately and loads selected patches and diagnostics
+  progressively. Read-only review stays available during work.
+- Filter files, navigate hunks with N/P, open a file in the configured editor with O, toggle
+  diagnostics with V, and refresh with Ctrl+R. Compact terminals reserve space for the patch.
+- Reviews show capture timestamps and identify truncated previews. Revert checks the captured
+  file, index, and HEAD again after approval; active work blocks revert and undo.
+- Completion receipts report recorded check-command exit statuses, separate from editor
+  diagnostics. Changed-file observations include Git-visible shell edits and journaled edits;
+  undo is offered only when its preflight succeeds and covers the observed files.
+- Command history retains 512 results. Unknown or compound shell commands are not classified as
+  individual checks, and missing results are reported explicitly.
+
+## Reliability and permission fixes
+
+- Repository configuration can no longer enable YOLO or override trusted permission, sandbox,
+  hook, credential, or checkpoint settings. Repository mode may tighten execution to plan mode.
+- Command classification requires approval for mutating or unrecognized filter options.
+  Broker checks validate the complete operation path and protect capabilities from replacement.
+- Checkpoint restore preserves protected files and refuses conflicting paths before writing.
+  Capture batches Git operations and bypasses repository content filters.
+- Interrupted-write recovery saves the current contents before restoring the preimage, preserving
+  repairs made after a crash. Empty files retain their correct undo/redo state after resume.
+- Journal retention is consistently bounded to 20 turns. Compaction artifacts preserve complete
+  mixed tool results, and live summarizers survive session reopening and older snapshots.
+- Timed-out hooks terminate their process groups and retain bounded output. Configuration writes
+  use TOMLKit to preserve valid TOML structure and comments while updating settings atomically.
+- Remove duplicate configuration writers, command registration, sampling helpers, and UI state.
+
+## Upgrade
+
+```bash
+noah update
+```
+
+Move privileged settings previously stored in repository configuration to trusted user
+configuration. See the [configuration guide](https://github.com/skundu42/noah-code/blob/v0.6.0/docs/configuration.md) and
+[updated keyboard reference](https://github.com/skundu42/noah-code/blob/v0.6.0/docs/interactive-reference.md).
+
+Local validation: 960 tests passed on Python 3.13/macOS, with one opt-in network test deselected;
+lint, type checks, lockfile validation, and wheel/source builds passed. The release workflow also
+validates package metadata and binary-only installation before publishing.
+
+**Full changelog:** https://github.com/skundu42/noah-code/compare/v0.5.1...v0.6.0

noah_code-0.7.0/docs/releases/v0.7.0.md

@@ -0,0 +1,51 @@
+# Noah Code v0.7.0
+
+Slash commands now run with one Enter or click. Check results are shared across agents and
+background jobs, and headless runs expose structured outcomes for automation.
+
+## Commands and editing
+
+- Enter or a single click runs the highlighted slash command or opens its picker. Tab completes
+  the command for editing; required arguments and file mentions remain editable.
+- Ctrl+P opens Noah's command picker and runs the selected command immediately. Displaced drafts
+  remain available through Alt+Z, and typed command arguments are preserved.
+- Prompt editing and clipboard handling are more reliable, including selection and paste behavior.
+
+## Shared verification
+
+- Foreground checks, streamed commands, subagents, background jobs, and terminal commands use a
+  shared session ledger. Records include the command, source, exit code, execution directory when
+  known, and workspace fingerprints before and after execution.
+- Receipts mark results stale when workspace files change during or after a check. Interrupted
+  checks remain incomplete, and unavailable workspace fingerprints are reported as unknown.
+- Agents can inspect the same evidence with `await self.ws.checks()`. Records survive session
+  reopening and are included in JSON outcomes. Headless outcomes refresh after process cleanup.
+- Workspace fingerprints use file metadata. Ignored untracked artifacts, external dependencies,
+  and metadata-preserving changes are outside their scope. Compound shell commands are not
+  credited as individual checks.
+
+## Task outcomes
+
+- Subagents requesting input retain a distinct `needs_input` state. The interface reports
+  "Input needed" instead of showing a completed turn.
+- `noah run --json` emits one result on stdout and sends progress to stderr. The result includes
+  `status`, `exit_code`, `explanation`, `session_id`, `run_id`, session `usage`, and `checks`.
+- Status distinguishes `completed`, `needs_input`, `failed`, and `cancelled`. Existing exit codes
+  remain compatible: both completion and waiting for input return zero, so automation should
+  inspect `status`. Task completion is separate from passing checks.
+- Headless follow-ups using `--session` continue the waiting run. Invalid agent stop reasons
+  and persistence failures are reported as failures.
+
+## Upgrade
+
+```bash
+noah update
+```
+
+Example structured run:
+
+```bash
+noah run --json "Fix the failing unit test" .
+```
+
+**Full changelog:** https://github.com/skundu42/noah-code/compare/v0.6.0...v0.7.0

{noah_code-0.5.1 → noah_code-0.7.0}/docs/reliability.md

@@ -53,16 +53,19 @@ fresh process-status snapshot, and continues the same journaled turn.
 
 Direct workspace writes and atomic multi-file patches record their pre-images in the runtime
 database before changing the checkout. A crash before the commit record restores those pre-images
-on the next launch. This protection complements the persistent edit journal used by `/undo` and
-`/redo`.
+on the next launch. Before rollback, the current contents and path metadata are saved in the session's
+`recovery/` directory, whose location is reported at startup. This preserves repairs made after the
+crash. If the backup cannot be saved, recovery stops before replacing that file. The persistent edit
+journal used by `/undo` and `/redo` retains the newest 20 turns in memory and across restarts.
 
 Git checkpoints are enabled by default. Noah captures them at turn boundaries and before mutating
 shell commands, stores them under `refs/noah-code/checkpoints/<session>/`, and keeps the newest 50
 by default. Capturing uses a temporary Git index and does not move `HEAD` or disturb the user's
-index. Capture is filter-free plumbing (`hash-object --no-filters` + `update-index --cacheinfo`),
+index. Capture batches filter-free hashing and index updates,
 so repository clean filters never execute during capture, and paths the permission engine
 classifies as secrets (`.env`, key stores, credential files) are never staged into checkpoint
-refs. Inspect and restore them with:
+refs. Restore also excludes these paths, preserving their current contents and index entries.
+Inspect and restore checkpoints with:
 
 ```text
 /checkpoints
@@ -139,6 +142,28 @@ unbounded growth. Delete sessions that are no longer needed with `noah sessions
 
 ## Observability
 
+### Verification evidence
+
+Foreground and streamed check commands, subagents, background jobs, and persistent-terminal
+commands share a bounded verification ledger. The agent can inspect it with
+`await self.ws.checks()`. Each record includes the observed exit code, command, source, execution
+directory when known, and workspace fingerprints before and after execution. Records persist with
+the session and are included in `noah run --json` results.
+
+Turn receipts recompute check status against the current workspace. A check becomes `stale` if
+files changed during or after its execution; missing fingerprints produce `unknown`, and an
+interrupted check is `incomplete`. A successful rerun supersedes the earlier attempt in the receipt.
+Completion and verification are separate: `completed` means the agent finished its turn, while
+the check records show the evidence available for its changes.
+
+Fingerprints use file metadata and symlink targets without reading file contents. In Git checkouts
+they cover tracked files and non-ignored untracked files; outside Git, common cache, build, and
+vendor directories are excluded. Ignored artifacts, external dependencies, and metadata-preserving
+edits are outside this check. Commands containing shell chaining, redirection, or expansion are
+not credited as individual checks because their exit code cannot establish each check's result.
+
+### Runtime status
+
 Use `/health` during idle or active work. It reports runtime database and artifact sizes, pending
 inbox items and interactions, live jobs, retained runtime events, and the current non-terminal run.
 Use `/status` for workspace/session state, `/tokens` for model and cache usage, `F2` for activity,

{noah_code-0.5.1 → noah_code-0.7.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "noah-code"
-version = "0.5.1"
+version = "0.7.0"
 description = "Noah Code terminal coding agent, built on NVIDIA OO Agents (NOOA)"
 readme = "README.md"
 requires-python = ">=3.12,<3.14"
@@ -33,6 +33,7 @@ dependencies = [
     "packaging>=24.0,<27.0",
     "rich>=13.0.0,<16.0.0",
     "textual>=1.0.0,<9.0.0",
+    "tomlkit>=0.13.2,<1.0.0",
 ]
 
 [project.urls]

{noah_code-0.5.1 → noah_code-0.7.0}/src/noah_code/__init__.py

@@ -1,3 +1,3 @@
 """Noah Code: terminal coding agent on NVIDIA OO Agents."""
 
-__version__ = "0.5.1"
+__version__ = "0.7.0"