Skip to content

docs: add Auggie 1.x preview & migration guide - #176

Open
augmentmoogi wants to merge 4 commits into
mainfrom
augmentmoogi/docs-v2-notes
Open

augmentmoogi wants to merge 4 commits into
mainfrom
augmentmoogi/docs-v2-notes

Conversation

@augmentmoogi

@augmentmoogi augmentmoogi commented Oct 12, 2026 •

Copy link
Copy Markdown
Contributor

Problem

Users need clear documentation explaining the upcoming Auggie 1.x release, distribution tags, breaking changes, and migration details.
Linear ticket: AGT-1142

Solution

  • Added docs/1.x-preview-guide.md covering distribution channels (latest, preview, daemon), feature highlights, breaking changes, WIP roadmap, and tag changelog.
  • Added link to the preview guide in README.md.

Pull Request opened by Augment Code | View session

@augmentmoogi augmentmoogi self-assigned this Oct 12, 2026
@augment-app-staging

Copy link
Copy Markdown
Contributor

PR Dashboard Manager 📊

Status Last run Actions
🐛 Deep Review No comment yet — cosmos review — Update review to head
🛡️ Risk Analysis No comment yet —
🧭 Pair Review No comment yet —
🔧 Fixer No comment yet — Start Fixer or cosmos fix
🧪 Verification No comment yet — cosmos verify — Rerun verifier
👤 Human Review 0 unresolved (0 total) —
How does this work?

Specialized experts work independently on the same change. The dashboard gathers their latest evidence in one place; humans remain responsible for judgment and merge.

🤖 Expert roles

  • 📊 PR Dashboard Manager — Observes the review lifecycle and keeps this dashboard current without controlling the specialists.
  • 🐛 Deep Code Reviewer — Reviews the diff line by line for objective, actionable bugs with minimal false positives.
  • 🛡️ PR Risk Analyzer — Classifies risk and approves only when the current head satisfies the configured policy.
  • 🧭 Pair Reviewer — Prepares a briefing on design, architecture, security, and testing to guide human review.
  • 🔧 PR Fixer — Handles actionable review feedback, failing checks, and merge conflicts after review begins.
  • 🧪 Verifier — Exercises applicable runtime behavior end to end and reports evidence for a human to judge.

👤 Human role

  • Review the Pair Review briefing and make judgement calls on highlighted design decisions.
  • Check Verifier evidence—or test manually when Verifier is unavailable.
  • Review Fixer comment replies to see what changes it made.
  • Make the final merge decision and merge the change; no expert merges it.

@augmentmoogi

Copy link
Copy Markdown
Contributor Author

PR Author Agent⚡

👋 I've got this PR.
I'll handle review feedback, CI failures, and merge conflicts.

@augmentmoogi
augmentmoogi marked this pull request as ready for review October 12, 2026 23:36
Comment thread docs/1.x-preview-guide.md
- **GitHub PR Reviews**: Preserved inline review comment reply relationships and threading.

#### `v1.2.0-prerelease.202610051436`
- **System Prompts**: Honored user-defined custom system prompts in session agent loops.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is not customer facing, internal only

@augment-app-staging

augment-app-staging Bot commented Oct 12, 2026 •

Copy link
Copy Markdown
Contributor

PR Risk Analyzer Agent 🛡️

👀 Human review needed

Justification
  • Reviewed the complete two-file diff: a README link and a new 165-line preview/migration guide. No executable code changes. The root approval policy is absent at base commit 9cc3ead419db9486ad44e6e4bba30ecd6784ccff, so the baseline policy applies.
  • The guide's “Hook Tool Names Changed” section promises compatibility for legacy hook matchers and permission rules. In the inspected Auggie 1.x source (augmentcode/cosmos-agent commit 9c46fe19d7e595390497c7ad221f2f6425a54184), aliases such as launch_process, read_file, and file_search are hook-only; permission matching uses a narrower shared alias list. Users could incorrectly assume an old permission restriction still applies. A release owner should verify and qualify that promise against the advertised preview builds, separating hook compatibility from permission compatibility.
  • Documentation-only scope does not remove this security-relevant migration uncertainty. Automatic approval is withheld; no builds or tests were run (static analysis only).

Reviewed commit d4f149c

Pair Reviewer 🧭

→ Pair Review Briefing

Reviewed commit d4f149c

@augment-app-staging augment-app-staging Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Deep Code Review Agent🐛

Review completed with 6 suggestions.

Reviewed commit: d4f149c

Comment thread docs/1.x-preview-guide.md
- **Snappier Terminal UI (Direct ANSI Engine)**:
In 0.x, the terminal interface could stutter or lag during fast token streaming or when printing large diffs. In 1.x, the rendering engine has been rewritten from scratch with a direct terminal renderer (`pi-tui`), delivering instant scrolling, smooth rendering, and zero redraw lag.
- **Native Editor Integration via ACP (Agent Client Protocol)**:
Auggie 1.x includes native support for ACP, enabling you to use Auggie as your coding agent directly inside editors like VS Code and Zed. Run `auggie acp` to start the protocol bridge and connect your favorite editor seamlessly.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Deep Code Review Agent🐛

auggie acp does not start the protocol bridge: parseArgs treats the bare acp argument as prompt text, leaving ACP mode disabled. An editor launched with this command cannot establish its ACP connection because it receives the normal CLI behavior instead of the protocol server.

Severity: medium


🤖 Was this useful? React with 👍 or 👎

Comment thread docs/1.x-preview-guide.md
- **Tool Permissions**: Restrict or require user confirmation for specific tools via `/permissions` or `.augment/settings.json`.

### Hook Tool Names Changed (With Temporary Backward Compatibility)
Tool names have been standardized to short, canonical verbs across Auggie 1.x. If you have custom hooks configured in `.augment/hooks.json` or plugin settings, note the following canonical tool name updates:

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Deep Code Review Agent🐛

.augment/hooks.json is not a project hook configuration source in 1.x: loadHooksConfigWithDiagnostics reads the hooks property from scoped settings files, while standalone hooks/hooks.json files are loaded only for plugins. Readers who configure or migrate their project hooks at the documented path will find that none of those hooks execute.

Severity: medium


🤖 Was this useful? React with 👍 or 👎

Comment thread docs/1.x-preview-guide.md
| `find` | `file-search`, `file_search`, `find-by-name` | Find files matching glob patterns |
| `bash` | `launch-process`, `launch_process`, `shell` | Run shell commands |

> **Note on Compatibility**: Temporary backward compatibility is provided in 1.x. Existing hook matchers and permission rules specifying legacy tool names (e.g. `matcher: "str-replace-editor"`) will continue to trigger for the canonical 1.x tools during the preview period. However, you should update your configuration to use the canonical names.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Deep Code Review Agent🐛

The listed aliases are not all supported by permission rules: names such as read-file and launch_process are hook-only aliases, whereas isRuleMatch checks the narrower shared getToolNameAliases set. Relying on this compatibility assurance can leave a deny rule ineffective for the canonical tool even though its corresponding hook still runs.

Severity: medium


🤖 Was this useful? React with 👍 or 👎

Comment thread docs/1.x-preview-guide.md
- **Improved TUI Navigation & Unified Key Mapping**:
Navigation across menus and popovers is now unified:
- **Type-to-Filter**: All selectors (model picker, worktree selector, settings) support typing directly to filter results.
- **Consistent Keybindings**: Standardized navigation across the interface (`j`/`k` or arrow keys, readline shortcuts like `Ctrl+A`/`Ctrl+E`, `Escape` to step back or dismiss).

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Deep Code Review Agent🐛

j and k are not default selector-navigation bindings: tui.select.up/down use arrow keys, and the model and worktree selectors pass those letters to their search inputs. Following this shortcut guidance filters the available choices instead of moving the selection.

Severity: low


🤖 Was this useful? React with 👍 or 👎

Comment thread docs/1.x-preview-guide.md
- **In-Session Inspectors (`/rules`, `/hotkeys`)**:
Inspect active configuration layers in real time: `/rules` to view loaded system and workspace rules, and `/hotkeys` for a complete keyboard shortcut reference.
- **Configurable Themes**:
Settings in `.augment/settings.json` now support custom color themes via `theme` (`"dark"`, `"light"`, or custom theme files) and persistent reasoning budgets via `defaultThinkingLevel`.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Deep Code Review Agent🐛

defaultThinkingLevel is not honored when starting a fresh session with an Augment-backed model: createAgentSession reads its model-specific reasoningEffortPreferences or uses the model default instead. Users who configure the advertised setting for their normal Augment sessions therefore will not get the persistent reasoning effort described here.

Severity: medium


🤖 Was this useful? React with 👍 or 👎

Comment thread docs/1.x-preview-guide.md

## 2. Feature Highlights (What’s New in 1.x)

> *Note: These are curated highlights of user-facing capabilities introduced in 1.x that were not available in 0.x. For release-by-release details, see the [Changelog](#5-detailed-changelog) below.*

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Deep Code Review Agent🐛

Several following highlights were already available in 0.x: this repository's CHANGELOG.md records ACP as fully released in 0.9.0, /rules improvements in 0.5.6, and interactive permission editing in 0.5.1. Calling all of these capabilities unavailable in 0.x misrepresents the migration benefits and implies users need a preview upgrade for existing stable features.

Severity: low


🤖 Was this useful? React with 👍 or 👎

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants