roborev
Marketplace for roborev agent skills.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Check it deeper on your machine
It has changed since its scan on 11 Nov 2026. Its scripts were only partly scanned.
As a Claude Code plugin, install it and then ask Claude Code:
/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/kenn-io/roborev with scanskillsafety
Or from a terminal, clone it and run it:
git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/kenn-io/roborev'
It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.
Badge for your README
Markdown
[](/listings/kenn-io-roborev)HTML
<a href="/listings/kenn-io-roborev"><img src="/api/listings/kenn-io-roborev/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
cd your-repo roborev init # Install post-commit hook git commit -m "..." # Reviews happen automatically roborev tui # View reviews in interactive UI roborev ui # Open reviews in the native browser UI
If roborev is managed by a version manager, `roborev init` and `roborev agent hook install` try to install hooks with the stable shim/symlink. You can also choose the exact binary path with `roborev init --binary ~/.local/share/mise/shims/roborev`, or `roborev agent hook install --binary ~/.local/share/mise/shims/roborev`.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 20 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.
- {
- "plugins": [
- {
- "version": "0.75.0",
- "name": "roborev",
- "description": "Agent skills for roborev code review, fix, refine, and response workflows.",
- "source": "./",
- "author": {
- "name": "roborev",
- "url": "https://github.com/roborev-dev"
- }
- }
- ],
- "name": "roborev",
- "description": "Marketplace for roborev agent skills.",
- "owner": {
- "name": "roborev",
- "url": "https://github.com/roborev-dev"
- }
- }
README.md · README · 644 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.
- ---
- last_edited: 2026-11-08
- ---
- <p align="center">
- <picture>
- <source media="(prefers-color-scheme: dark)" srcset="https://roborev.io/docs/assets/static/logo-with-text-dark-bg.svg">
- <img alt="roborev" src="https://roborev.io/docs/assets/static/logo-with-text-light.svg">
- </picture>
- </p>
- <p align="center">
- <a href="https://go.dev/"><img src="https://img.shields.io/badge/Go-1.27.4+-00ADD8?logo=go&logoColor=white" alt="Go 1.27.4+"></a>
- <a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="License: MIT"></a>
- <a href="https://roborev.io"><img src="https://img.shields.io/badge/Docs-roborev.io-blue" alt="Docs"></a>
- </p>
- <p align="center">
- <a href="https://roborev.io/docs/">Documentation</a> ·
- <a href="https://roborev.io/docs/quickstart/">Quick Start</a> ·
- <a href="https://roborev.io/docs/installation/">Installation</a>
- </p>
- **Continuous code review for AI coding agents.**
- roborev runs in the background, reviews every commit as agents write code,
- and surfaces issues in seconds -- before they compound. Pull code reviews
- into your agentic loop while context is fresh.
- <p align="center">
- <img src="https://roborev.io/docs/assets/generated/web-ui.png" alt="Roborev browser application">
- </p>
- ## How It Works
- 1. Run `roborev init` to install a post-commit hook
- 2. Every commit triggers a background review -- agents write, roborev reads
- 3. View findings in the TUI, feed them to your agent, or let `roborev fix` handle it
- ### Automation, two layers
- 
- - **Post-commit reviews** - a git hook reviews every commit in the background (any agent).
- - **Agent hook** - watches supported coding-agent sessions and brings open
- roborev findings back into the active workflow.
- ```bash
- roborev init # layer 1: per-commit reviews
- ```
- Not in the latest release yet
- ```bash
- roborev agent setup # layer 2: wire Claude Code and Codex
- ```
- For other supported agents, run `roborev agent hook install --agent all`.
- Before you ship, run the `/roborev-refine` skill: it re-reviews and fixes your
- whole branch until every review passes, catching bugs before the PR.
- New here? Run `roborev quickstart` and point your agent at it.
- Not in the latest release yet
- Add `--mcp` to `roborev agent setup` to register MCP. If Codex asks you to
- review hooks, run `/hooks` in Codex, then `roborev agent setup --check`.
- ## Quick Start
- ```bash
- cd your-repo
- roborev init # Install post-commit hook
- git commit -m "..." # Reviews happen automatically
- roborev tui # View reviews in interactive UI
- roborev ui # Open reviews in the native browser UI
- ```
- If roborev is managed by a version manager, `roborev init` and
- `roborev agent hook install` try to install hooks with the stable shim/symlink.
- You can also choose the exact binary path with
- `roborev init --binary ~/.local/share/mise/shims/roborev`, or
- `roborev agent hook install --binary ~/.local/share/mise/shims/roborev`.
- 
- ## What you can do
- - **Background Reviews** - Every commit is reviewed automatically via
- git hooks. No remote review workflow required.
- - **Auto-Fix** - `roborev fix` feeds review findings to an agent that
- applies fixes and commits. `roborev refine` iterates until reviews pass.
- - **Agent Hook** - Optional hooks for Claude Code, Codex, Copilot CLI, Cursor,
- Factory Droid, Gemini CLI, Hermes, and Qwen bring open findings back into the
- active agent session.
- - **Custom Reviews** - Write your own rubrics and run them as review panels,
- for code or for documents such as legal drafts. `/roborev-setup-review`
- interviews you and writes the setup. Reviewers can read PDFs that a commit
- adds or changes, and strict review types get only read tools plus the MCP
- tools you grant.
- - **Code Analysis** - Built-in analysis types (duplication, complexity,
- refactoring, test fixtures, dead code, security) that agents can fix
- automatically.
- - **Multi-Agent** - Works with Codex, Claude Code, Gemini, Copilot,
- OpenCode, Cursor, Kiro, Kilo, Droid, Pi, and Muse Code.
- - **Runs Locally** - No hosted service or additional infrastructure.
- Reviews are orchestrated on your machine using the coding agents
- you already have configured.
- - **Interactive TUI** - Real-time review queue with vim-style navigation.
- Large terminals show the queue and selected review in a split-screen layout.
- - **Native Browser UI** - Browse, filter, inspect, comment on, close, cancel,
- and rerun reviews from the application embedded in the local daemon, then
- explore cost, latency, reliability, and outcomes in Analytics.
- - **Review Verification** - `roborev job compact` verifies findings against
- current code, filters false positives, and consolidates related issues
- into a single review.
- - **Review History Search** - Search completed reviews across repositories by
- keyword or meaning. Lexical search stays local and needs no embedding
- provider; semantic and hybrid modes are optional.
- - **Extensible Hooks** - Run shell commands on review events. Built-in
- [beads](https://github.com/steveyegge/beads) and [kata](https://github.com/kenn-io/kata)
- integrations create trackable issues from review failures automatically.
- ## The Agentic Fix Loop
- When reviews find issues, copy-and-paste the reviews into your
- interactive agent sessions, or invoke `/roborev-fix` in Claude Code
- or `$roborev-fix` in Codex. You can also address open reviews on the
- command line non-interactively with `roborev fix`.
- `roborev fix` shows the review findings to an agent, which applies
- changes and commits. The new commit gets reviewed automatically,
- closing the loop.
- `roborev agent hook install` auto-detects installed Claude Code, Codex, Copilot
- CLI, Cursor, Factory Droid, Gemini CLI, Hermes, Qwen, Grok Build, and ZCode
- harnesses
- and adds optional hooks after configured turn, commit, or failed-review
- thresholds are met.
- Reminders name exact review IDs and invoke the bundled `roborev-fix` skill.
- <!-- release-commands: expected-failure -->
- Never run `roborev fix --open`.
- Hook installation updates bundled skills for every supported profile. Hermes delivers queued
- post-tool reminders at `Stop`; Cursor records the same events but emits no control
- response.
- Installed hooks post events to the regular roborev daemon. That daemon evaluates
- the reminders and persists session counters and delivered review IDs in
- `${ROBOREV_DATA_DIR:-~/.roborev}/agent-hook/state.json`. Hook callbacks fail open
- when the daemon is unavailable, so they do not block the coding agent.
- For fully automated iteration (advanced feature), use `refine`:
- ```bash
- roborev refine # Fix, re-review, repeat until passing
- roborev refine --plan # Plan each fix before editing
- roborev refine --plan-only # Inspect plans for existing failed reviews
- roborev refine --skip-trivial-rereview # Do not wait on docs/test-only fix re-reviews
- ```
- `refine` runs in an isolated worktree and loops: fix findings, wait for
- re-review, fix again, until all reviews pass or `--max-iterations` is hit.
- ## Rebutting a Finding
- Use a stored reviewer session to re-evaluate one finding with your explanation:
- ```bash
- roborev job comment 42 --rebuttal --finding "Missing null check in handler." \
- -m "The caller rejects null input before invoking the handler."
- ```
- The assessment is saved as a comment. Add `--close` to close the whole review
- once it is recorded. See
- [Responding to Reviews](https://roborev.io/guides/responding-to-reviews/) for
- supported agents and recovery.
- ## Code Analysis
- Run targeted analysis across your codebase and optionally auto-fix:
- ```bash
- roborev analyze duplication ./... # Find duplication
- roborev analyze refactor --fix *.go # Suggest and apply refactors
- roborev analyze complexity --wait main.go # Analyze and show results
- roborev analyze test-fixtures *_test.go # Find test helper opportunities
- roborev analyze security ./... # Find security risks in existing code
- ```
- Available types: `test-fixtures`, `duplication`, `refactor`, `complexity`,
- `api-design`, `dead-code`, `architecture`, `security`.
- Analysis jobs appear in the review queue. Use `roborev fix` to apply
- open findings later, target a specific job with `roborev fix <id>`, or
- pass `--fix` to apply immediately.
- Analysis types can pin their own agent settings in config:
- ```toml
- [analyze.refactor]
- agent = "claude-code"
- model = "sonnet"
- reasoning = "fast"
- ```
- ## Installation
- **macOS ([Homebrew](https://formulae.brew.sh/formula/roborev)):**
- ```bash
- brew install roborev
- ```
- Homebrew manages updates with `brew upgrade roborev`.
- **Linux (install script):**
- ```bash
- curl -fsSL https://roborev.io/install.sh | bash
- ```
- Linux users can also install a [DEB or RPM package](https://roborev.io/docs/installation/#linux-packages-deb-and-rpm).
- **Windows (PowerShell):**
- ```powershell
- powershell -ExecutionPolicy ByPass -c "irm https://roborev.io/install.ps1 | iex"
- ```
- **With Go:**
- ```bash
- go install go.kenn.io/roborev/cmd/roborev@latest
- ```
- The Go module source archive does not include generated browser assets. This
- installation path provides the CLI and terminal UI; install a release package
- or build with `make install` for the embedded browser application.
- ## Developer Setup
- This repo uses [`prek`](https://prek.j178.dev/) for local Git hooks. Commits
- run only the Renovate config validator and jactionlint, and only when
- `renovate.json` or a workflow file changes. The whole-repo checks run on push:
- a Git-test isolation guard, `make lint-ci` for non-mutating Go lint,
- `make markdown-ci` for Zensical Markdown formatting, and the web check and test
- scripts when web files change. If you installed the hooks before, rerun
- `prek install` to add the pre-push hook.
- ```bash
- brew install prek uv # or use your preferred install method
- mise use --global npm:renovate@latest
- prek install # install the pre-commit and pre-push hooks
- prek run --all-files --hook-stage pre-push # run the push checks manually
- ```
- Use `make lint` when you explicitly want golangci-lint to apply fixes. Use
- `make markdown` to wrap prose in published Zensical pages at 80 columns while
- leaving Markdown tables unchanged. Use `make check-renovate-config` to validate
- `renovate.json` directly.
- ## Commands
- | Command | Description |
- |---------|-------------|
- | `roborev init` | Initialize roborev in current repo |
- | `roborev tui` | Interactive terminal UI |
- | `roborev daemon status` | Show daemon, browser UI, and queue status |
- | `roborev review <sha>` | Queue a commit for review |
- | `roborev review --branch` | Review all commits on current branch |
- | `roborev review --dirty` | Review uncommitted changes |
- | `roborev fix` | Fix open reviews (or specify job IDs) |
- | `roborev refine` | Auto-fix loop: fix, re-review, repeat |
- | `roborev analyze <type>` | Run code analysis with optional auto-fix |
- | `roborev agent hook install` | Install hooks for detected coding agents |
- | `roborev agent hook install --agent all` | Install all nine supported integrations |
- | `roborev agent hook snooze` | Silence Agent Hook reminders in the current worktree and branch |
- | `roborev agent hook snooze off` | Resume Agent Hook reminders in the current worktree and branch |
- | `roborev job compact` | Verify and consolidate open review findings |
- | `roborev job show [sha]` | Display review for commit |
- | `roborev job search <query>...` | Search completed review history across repositories |
- | `roborev export reviews` | Export completed reviews as JSON |
- | `roborev export ci-metrics` | Export finalized CI panel metrics as JSON |
- | `roborev export ci-costs` | Export job-level CI costs as JSON |
- | `roborev run "<task>"` | Execute a task with an AI agent |
- | `roborev job close <id>` | Close a review |
- | `roborev agent skills install` | Install bundled skills for supported agent profiles |
- See [full command reference](https://roborev.io/docs/commands/) for all options.
- Review history search is global by default and supports repository, branch,
- time, verdict, and open/closed filters. Lexical search works with no provider
- configuration or network call. See the
- [search guide](https://roborev.io/docs/search/) for semantic and hybrid setup,
- freshness, privacy, and recovery behavior.
- ### Exporting review history
- Use `roborev export reviews` to emit completed reviews as one JSON document for
- local reporting or archival workflows:
- ```bash
- roborev export reviews
- roborev export reviews --profile metadata --since 2026-06-01 --until 2026-06-30
- roborev export reviews --closed-only --repo github.com/org/repo --limit 1000
- roborev export reviews --cursor "$NEXT_CURSOR" --until 2026-07-01
- ```
- The default `content` profile includes each review twice: `document` is the
- stored JSON review document with the summary, verdict, and findings, and
- `content` is the Markdown rendering of it. Review text may contain sensitive
- repository details, so handle exported files carefully.
- Use `--profile metadata` when you only need identifiers, timestamps, verdicts,
- cost metadata, and related review metadata.
- Exports include a stable `database_id` for the local review database and, when
- at least one review is emitted, an opaque `next_cursor`. Pass
- `--cursor <next_cursor>` to resume after the previous page; `--cursor` cannot
- be combined with `--since`. If a cursor belongs to a previous database
- generation, `roborev export reviews` exits with code `3`; discard the cursor
- and retry with a window backfill. Other cursor rejections also require
- discarding the cursor before backfilling.
- Each exported review reports `closed` and `updated_at`. Closing or reopening a
- review does not move it past a `completed_at` cursor, so combine the cursor pull
- with `roborev export reviews --updated-since <time>` to pick up later close and
- reopen changes. See the
- [export reference](https://roborev.io/docs/commands/#exporting-reviews).
- Use `roborev export ci-metrics` to emit finalized CI panel runs — terminal
- outcome (`review_posted`, `review_suppressed`, `no_review_posted`,
- `giveup_posted`, `abandoned`, or `unknown` for panels finalized before outcomes
- were recorded), first-attempt and posting timestamps, attempt count, and each
- panel's member/synthesis jobs — for external review turnaround tracking.
- `giveup_posted` is a historical outcome; exhausted retries now finish with
- `no_review_posted` and an error status without a PR comment. The export follows
- the same cursor contract as `roborev export reviews`, ordered by `posted_at`,
- and exits with code `3` when a cursor's `database_id` no longer matches so
- callers can discard the cursor and backfill.
- Pass `--legacy` to export the frozen pre-panel CI era instead (rows with
- outcome `legacy_review`, one per reviewed PR head, from before panel runs
- existed) as a one-time backfill; legacy and panel cursors are namespaced
- and cannot be resumed against each other's export.
- Use `roborev export ci-costs` to emit cost-eligible CI jobs, including terminal
- retry attempts that are no longer retained by a panel. Each row records its
- completion time, agent, panel role, terminal status, and estimated USD cost.
- Jobs whose agent ran but whose model cannot be priced remain present with
- `cost_usd: null`; a reported free run is represented as `0`.
- Cost rows are ordered by `(finished_at, job_id)` for stable pagination. A fresh
- export over an overlapping window returns the current price for every matching
- job, so idempotent consumers can pick up pricing recorded after an earlier
- export. The export follows the same database-reset and opaque-cursor contract
- as the other exports. `--legacy` selects the structurally identified pre-panel
- CI era for a one-time historical backfill; regular and legacy cost cursors
- cannot be mixed.
- ## Configuration
- Create `.roborev.toml` in your repo:
- ```toml
- agent = "claude-code"
- snapshot_dir = ".roborev"
- review_guidelines = """
- Project-specific review instructions here.
- """
- # Optional: use repo guidelines instead of appending global review_guidelines.
- review_guidelines_supersede_global = false
- # Optional: metadata for roborev-owned fix commits and prompt hints for agent-owned fix commits.
- fix_commit_author = "Your Name <you@example.com>"
- fix_commit_co_authored_by = ["Pair Reviewer <pair@example.com>"]
- ```
- You can also set `review_guidelines` in `~/.roborev/config.toml`. Global
- guidelines apply to every repo and are appended before repo guidelines by
- default.
- `snapshot_dir` must be repo-relative. `roborev init` ensures it is ignored in `.gitignore`; snapshot creation also adds a local `.git/info/exclude` fallback for existing checkouts whose ignore setup is stale.
- See [configuration guide](https://roborev.io/docs/configuration/) for all options.
- ### Kata task context
- If your repo is bound to a [kata](https://github.com/kenn-io/kata) project (a
- committed `.kata.toml`), roborev can pull the kata issue(s) referenced in the
- reviewed commit messages into the review prompt, and file review findings back
- as kata issues.
- ```toml
- # .roborev.toml
- [kata_context]
- mode = "current" # off (default) | current | open
- max_chars = 50000 # cap on kata context bytes in the prompt
- # File a kata issue when a review fails or returns findings:
- [[hooks]]
- event = "review.*"
- type = "kata"
- # branches = ["main"] # only file katas for reviews on these branches; default all
- # project = "myproj" # defaults to the .kata.toml binding
- # labels = ["from-review"]
- # priority = 2
- ```
- `mode = "current"` includes only the katas referenced anywhere in the reviewed
- commit messages (e.g. `Closes: kata#abc4`); `open` includes every open kata in
- the bound project, except issues the kata hook itself filed (labelled
- `roborev`), so review findings are not fed back into later reviews as task
- intent. A hook's optional `branches` list (glob patterns such as `release/*`)
- limits it to reviews on matching branches; unset fires on all branches. The
- matched branch is the commit's branch for local reviews and the PR base
- (target) branch for CI pull-request reviews — so `branches = ["main"]` means
- commits on `main` locally but PRs *targeting* `main` in CI, and a fork's head
- branch name cannot satisfy a protected-branch filter. The `kata` CLI must be
- on `PATH`; when it is absent or the repo is unbound, prompt context is
- silently skipped. Any other failure — a broken `.kata.toml`, a failing `kata`
- invocation — is logged by both the prompt builder and the configured `kata`
- hook, so a configured integration never goes dark unnoticed.
- ### Environment Variables
- | Variable | Description |
- |----------|-------------|
- | `ROBOREV_DATA_DIR` | Override default data directory (`~/.roborev`) |
- | `ROBOREV_TELEMETRY_ENABLED` | Set to `0` to disable anonymous usage telemetry |
- | `EDITOR` | Editor used by `roborev job comment` when no message is supplied |
- | `GH_TOKEN`, `GITHUB_TOKEN` | GitHub authentication for CI commands |
- | `GITLAB_TOKEN`, `GL_TOKEN` | GitLab authentication for CI commands |
- | `ROBOREV_COLOR_MODE` | TUI color theme: `auto` (default), `dark`, `light`, `none` |
- | `ROBOREV_SYNC_CURSOR_LOOKBACK` | PostgreSQL sync cursor overlap duration (default `5m`) |
- | `ROBOREV_AGENT_HOOK_TURN_THRESHOLD` | Override agent-hook Stop threshold |
- | `ROBOREV_AGENT_HOOK_COMMIT_THRESHOLD` | Override agent-hook commit threshold |
- | `ROBOREV_AGENT_HOOK_FAILED_REVIEW_THRESHOLD` | Override agent-hook failed-review threshold |
- | `ROBOREV_AGENT_HOOK_ROBOREV_ADDR` | Override the regular daemon address used by Agent Hook |
- | `ROBOREV_DROID_HOOK_TURN_THRESHOLD` | Override Factory Droid agent-hook Stop threshold |
- | `ROBOREV_DROID_HOOK_COMMIT_THRESHOLD` | Override Factory Droid agent-hook commit threshold |
- | `ROBOREV_DROID_HOOK_FAILED_REVIEW_THRESHOLD` | Override Factory Droid agent-hook failed-review threshold |
- | `NO_COLOR` | Set to any value to disable all color output ([no-color.org](https://no-color.org)) |
- ## Supported Agents
- | Agent | Install |
- |-------|---------|
- | Codex | `npm install -g @openai/codex` |
- | Claude Code | `npm install -g @anthropic-ai/claude-code` |
- | Gemini | `curl -fsSL https://antigravity.google/cli/install.sh \| bash` (preferred Antigravity CLI) or `npm install -g @google/gemini-cli` |
- | Copilot | `npm install -g @github/copilot` |
- | OpenCode | `npm install -g opencode-ai@latest` ([anomalyco/opencode](https://github.com/anomalyco/opencode)) |
- | Cursor | [cursor.com](https://www.cursor.com/) |
- | Kiro | [kiro.dev](https://kiro.dev/) |
- | Kilo | `npm install -g @kilocode/cli` |
- | Droid | [factory.ai](https://factory.ai/) |
- | Pi | [pi.dev](https://pi.dev/) |
- | Muse Code | [Meta installation guide](https://dev.meta.ai/docs/overview) |
- roborev auto-detects installed agents.
- To use Pi as the auto-design routing classifier (`classify_agent = "pi"`),
- install the JSON Schema output extension too:
- ```bash
- pi install npm:@nqbao/pi-json-schema
- ```
- roborev loads this extension explicitly when it invokes the classifier. Keeping
- it installed in Pi makes the classifier setup visible in `pi list` and avoids
- runtime package-fetch surprises in offline or locked-down environments.
- ### Routing Claude Code to a proxy (Ollama, LiteLLM, etc.)
- The `claude-code` agent accepts a model spec of the form `<model>@<base_url>`.
- When `<base_url>` starts with `http(s)://`, roborev points Claude Code at
- that endpoint and pins all tier aliases (Opus/Sonnet/Haiku/subagent) to the
- given model.
- ```toml
- # .roborev.toml — local Ollama for reviews, real Anthropic for fixes
- agent = "claude-code"
- review_model = "glm-5.1:cloud@http://127.0.0.1:11434"
- fix_model = "sonnet"
- ```
- Or via CLI: `roborev review --model 'glm-5.1:cloud@http://127.0.0.1:11434'`.
- **Proxy auth.** Set `ROBOREV_CLAUDE_PROXY_TOKEN` to forward a bearer token
- to the proxy as `ANTHROPIC_AUTH_TOKEN`. If unset, roborev sends a placeholder
- (sufficient for gateways that don't check the header, such as Ollama).
- roborev does *not* forward `ANTHROPIC_API_KEY` to proxy endpoints — that
- would leak a real Anthropic credential to arbitrary third parties.
- **URL restrictions.** Proxy URLs must not embed `user:pass@` credentials
- (use `ROBOREV_CLAUDE_PROXY_TOKEN`); `http://` is only accepted for loopback
- hosts (`127.0.0.1`, `::1`, `localhost`) so plaintext endpoints can't receive
- tokens over the wire. Use `https://` for remote proxies. The full URL
- (including any path or query string) is forwarded as-is to
- `ANTHROPIC_BASE_URL`, so include the path your gateway expects (e.g.
- LiteLLM may want a trailing `/v1`; Ollama wants no path).
- **Environment behavior (breaking change in this release).** When the
- `claude-code` agent runs, roborev always strips inherited `ANTHROPIC_API_KEY`,
- `ANTHROPIC_BASE_URL`, `ANTHROPIC_AUTH_TOKEN`,
- `ANTHROPIC_DEFAULT_{OPUS,SONNET,HAIKU}_MODEL`, and `CLAUDE_CODE_SUBAGENT_MODEL`
- from the child environment. If you were previously routing Claude Code by
- exporting these vars in your shell, switch to the `<model>@<base_url>` spec
- instead. For native (non-proxy) mode, configure `ANTHROPIC_API_KEY` via
- roborev's config (it is re-injected from roborev's stored key, not inherited
- from the operator's shell).
- ## Telemetry
- Release builds of roborev send limited anonymous telemetry to PostHog when the
- daemon starts and once every 24 hours while it remains running: `daemon_started`
- and `daemon_active` with `repo_count` and `review_count` as buckets: `0-5`,
- `6-20`, `21-100`, or `over-100`. These events also include sync enabled,
- CI enabled, and auto-design enabled, plus `application=roborev`, version, OS/arch,
- `$process_person_profile=false`, `$geoip_disable=true`, and an anonymous install
- ID.
- The daemon sends `setup_finished` once per install the first time it finds a
- finished review: one that ran locally, arrived through sync, or already existed
- when the daemon started. Panel member reviews count. The event has no
- properties. Its timestamp records when the daemon noticed the review, so it can
- come after an upgrade or after telemetry is turned on. If sending fails, the
- daemon does not retry.
- Closing a browser tab, leaving it hidden for 30 minutes, or exiting the TUI
- reports one `session_ended` with `surface` and `duration_bucket`: `under_1m`,
- `1_to_5m`, `5_to_30m`, or `over_30m`. Each browser session adds up visible time
- across tab switches and excludes hidden time. A browser session is lost when the
- browser quits, crashes, or discards the tab before the page closes normally.
- The web UI asks the daemon to report `app_opened` once per UTC day per browser
- profile, on load or focus. `roborev tui` asks each time it starts. CLI commands
- that work through the daemon ask after their first successful daemon request,
- without starting a daemon; hook, MCP and daemon management commands, and
- commands the bundled agent skills run with `--from-skill`, do not. Agent calls
- count separately as described below. The daemon sends at most one `app_opened`
- per surface per UTC day, so repeated opens on the same day count once, with the
- same install ID and default fields plus `surface` (`web`, `tui` or `cli`). The
- browser, the TUI and the CLI only talk to the daemon; the TUI and the CLI also
- send nothing when an opt-out variable is set in their own environment. The
- browser and TUI also report `screen_viewed` with a fixed `screen` name and
- `surface` of `web` or `tui`. Browser screens are `reviews` and `analytics`. TUI
- screens are `queue`, `review`, `prompt`, `filter`, `comment`, `commit-msg`,
- `help`, `log`, `tasks`, `worktree-confirm`, `patch`, `column-options`,
- `release-notes` and `rerun-agent`. Each screen counts once per install per UTC
- day across interfaces, including daemon restarts. Screen names contain no paths,
- job IDs, filter text or review contents. `review_viewed` counts a rendered
- review read once per installation, surface, and UTC day. It carries
- `surface=web|tui|cli`; the TUI requires key or mouse input before counting
- loaded content. `finding_fixed` counts each committed fix with
- `surface=tui|cli`, and `review_closed` counts each successful explicit close
- with `surface=web|tui|cli`. CLI actions require terminal output and exclude
- agents and `--from-skill`. Successful reopens and automatic closes stay silent.
- Anonymous telemetry also counts core action error banners and toasts in the web
- UI and TUI, and PostgreSQL sync outages. Failed closes and reopens both count.
- Daemon-unreachable errors can't be counted. See [telemetry configuration](https://github.com/kenn-io/roborev/blob/main/docs/configuration.md#telemetry)
- for event properties and reporting rules.
- `search_empty` counts each Enter-submitted browser search that returns no jobs
- and each human CLI search that finds no reviews, with only `surface=web|cli`.
- Previews and search text are excluded.
- Daily screen and `review_viewed` claims share `telemetry-daily.json` beside the
- daemon database, wherever `--db` points, with a `telemetry-daily.json.lock`
- sidecar. Deleting the file resets its claims; existing SQLite claims remain
- effective through their recorded UTC dates. See [telemetry
- configuration](https://github.com/kenn-io/roborev/blob/main/docs/configuration.md#telemetry)
- for recovery instructions. Counts mean the report was queued, not confirmed
- delivered to PostHog. The TUI retries a failed screen report on the next key or
- mouse input while that screen is showing. The browser sends each screen and
- review read once per UTC day per browser profile, across reloads and window
- focus returns. It keeps these daily telemetry claims in localStorage.
- After a `disabled` receipt, the browser sends nothing until a later UTC day's
- load or focus checks again. A TUI, CLI, or MCP process stops reporting until the
- next UTC day, a TUI reconnect, or its own restart. The agent-call route returns
- the same `disabled` or `queued` JSON receipt as the event route.
- Agent activity counts each MCP `tools/call`, including tool errors, and each
- invocation of `agent-hook run`, `agent-hook fix-done`, `post-commit`, `enqueue`,
- `remap`, or an eligible daemon command marked `--from-skill` after its first
- successful product request. MCP discovery and hooks that do no daemon work
- add nothing. The daemon emits `agent_active` once per install per UTC day with
- `call_count_bucket=1-10`, then `agent_call_count` only on calls 11 and 101 with
- `11-100` and `over-100`. The highest bucket observed that day represents its
- volume. Counts persist across restarts in the existing local database.
- Delivery is best effort and honors all opt-outs.
- CLI and stdio MCP notifications and TUI exit reports use a one-second request
- timeout. CLI commands, stdio MCP sessions, and the TUI wait up to 50 ms at exit
- for unfinished reports. Finished reports add no delay. An unfinished report may
- be lost when the process exits. Telemetry alone never starts a daemon.
- Each event also carries `install_age_hours`, the whole hours since the install
- was created, so short-lived installs such as test sandboxes can be filtered out.
- Each event also carries `distribution`, the channel that built the binary, such
- as `homebrew` or `kenn` for official direct downloads. It's empty for source
- builds.
- It does not send repo names, paths, remotes, prompts, review output, provider
- tokens, usernames, or IP geolocation. Set `ROBOREV_TELEMETRY_ENABLED=0`,
- `TELEMETRY_ENABLED=0`, or `DO_NOT_TRACK=1` to disable it, then restart the
- daemon. Telemetry is always off in CI. Release builds, `make build`, and
- `make install` set the PostHog project key at link time. Plain `go build`,
- `go install`, and Go tests have no key and send no telemetry.
- ## Security Model
- roborev delegates code review and fix tasks to AI coding agents that
- have shell access. Review agents may execute read-only git and shell
- commands to inspect diffs; fix agents run in isolated worktrees with
- full tool access. A custom review type with a strict tool policy runs
- Claude Code in a locked-down session with only read tools and the MCP
- tools the operator allows.
- **roborev is designed for use with trusted codebases.** The review
- prompt includes diff content and commit messages from the repository.
- If you are reviewing untrusted code (e.g., open-source contributions
- from unknown authors), run roborev inside a sandboxed environment
- (container, VM, or similar) to limit the blast radius of any
- prompt-injection attack that could cause an agent to execute
- unintended commands.
- ## Documentation
- Full documentation available at **[roborev.io](https://roborev.io)**:
- - [Quick Start](https://roborev.io/docs/quickstart/)
- - [Installation](https://roborev.io/docs/installation/)
- - [Commands Reference](https://roborev.io/docs/commands/)
- - [Configuration](https://roborev.io/docs/configuration/)
- - [Auto-Fixing with Refine](https://roborev.io/docs/guides/auto-fixing/)
- - [Code Analysis and Assisted Refactoring](https://roborev.io/docs/guides/assisted-refactoring/)
- - [Hooks](https://roborev.io/docs/guides/hooks/)
- - [Agent Hook](https://roborev.io/docs/agent-hook/)
- - [Agent Skills](https://roborev.io/docs/guides/agent-skills/)
- - [PostgreSQL Sync](https://roborev.io/docs/advanced/postgres-sync/)
- For local development in this repo, install hooks with `prek install` or run
- `make install-hooks` as a thin wrapper around `prek install`.
- ## License
- MIT
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.