Core Concepts
AI Hub Actions
Open the AI Hub — press a in the TUI or Cmd+K → AI (macOS) /
Ctrl+K → AI (Windows/Linux) in the desktop app — to run any of these
actions. Each writes sidecar files that er reads and renders inline. The Hub is the primary way to
trigger AI work; no external tool or command is needed.
The standard loop
| Step | AI Hub / App | Action | What happens |
|---|---|---|---|
| 1 | AI Hub | TUI: Review work · Desktop: Run review | Writes review sidecar files; findings appear inline |
| 2 | er |
A / Ctrl+j / Ctrl+k | Findings appear inline (toggle the layer with A); jump between findings with Ctrl+j / Ctrl+k — J/K move between comments and questions instead |
| 3 | er |
q / c | Leave questions and comments on lines |
| 4 | AI Hub (TUI only) | Answer questions |
Reads your questions, appends an AI answer to each, and marks them resolved. Desktop: use the Notes panel instead — there is no Hub action for this. |
| 5 | er |
— | Auto-refreshes; AI responses appear inline |
| 6 | Not a Hub action |
TUI: Git hub (g) → Push comments to GitHub Desktop: Comments card → Push to GitHub |
Posts your local comments to the GitHub PR as inline comments |
Full catalog
All actions are available in both apps unless noted. Where TUI and Desktop differ, the difference is called out explicitly.
Run review
| App | Invocation |
|---|---|
| TUI | Press a → select Review work |
| Desktop | Press Cmd+K → AI → Run review |
Full review: risk levels, findings by severity, review order, checklist, and a markdown summary.
Writes review.json, order.json, checklist.json, and summary.md.
Specialized / expert review
Eight expert lenses, each focused on a single dimension and writing only to
experts/<id>.json. The TUI and Desktop offer the same experts but through different UI flows.
TUI: Press a → Specialized review → a sub-menu lists all 8 experts; select one to run it.
Desktop: Press Cmd+K → AI → Run reviewers… → a
Choose reviewers multi-select appears (Triage + General + 8 experts + Professor); toggle with
Space and confirm with Run N reviewer(s). You can run multiple in one go.
| Expert | Focus | Writes |
|---|---|---|
| Security | Auth, injection, data exposure | experts/security.json |
| Performance | Allocation, hot paths, complexity | experts/performance.json |
| Reliability | Error handling, resilience | experts/reliability.json |
| Testing | Coverage, assertion quality, mocks | experts/testing.json |
| API / contracts | Interface stability, breaking changes | experts/api.json |
| Patterns | Consistency with existing codebase conventions | experts/patterns.json |
| Simplifying | Readability and complexity reduction | experts/simplifying.json |
| Mentorship | Exemplary patterns worth fostering | experts/mentorship.json |
Triage
| App | Invocation |
|---|---|
| TUI | Press a → Triage branch |
| Desktop | Press Cmd+K → AI → Triage branch |
Fast branch scan: first impression, a routing verdict on what to run next (General review, one or more
expert lenses, Arena debate, Professor, or skip deep review), and priority files. Writes
triage.json. Useful before committing to a full review.
Risk rules
| App | Invocation |
|---|---|
| TUI | Press a → Propose risk rules |
| Desktop | Settings → Review → File importance → Propose rules |
Reads the repo and writes this repo's [risk.<repo>] table into the global
config, where each rule declares a path, glob or file type as high, med,
low or ai (no fixed tier: the AI review rates the file). Rules you write
yourself under [user_risk] apply in every repo, after the repo's own. The tiers rank
files without asking a model at review time, so an unreviewed diff can be filtered with
importance:high. An older [importance.<repo>] table is still read
(foundational as high, isolated as low, normal as ai).
The agent reports the share each tier covers; a table that claims most of the tree is visible there.
Professor
| App | Invocation |
|---|---|
| TUI | Press a → Professor |
| Desktop | Press Cmd+K → AI → Professor (opens an optional focus-prompt modal before running) |
A learning agent — it teaches rather than critiques. Surfaces concepts, patterns, and insights from
the diff. Writes professor.json. See skills/PROFESSOR_PHILOSOPHY.md for the
philosophy behind it.
Generate summary
TUI only: Press a → Generate summary to regenerate
summary.md as a standalone action.
Desktop: There is no standalone "Generate summary" action in the ⌘K AI menu.
summary.md is produced automatically as part of Run review.
Answer questions
TUI only: Press a → Answer questions. This action is
disabled until you have an unresolved question (q adds one). It reads your
questions, generates AI responses, and updates questions.json.
Desktop: There is no Hub action for answering questions. Questions are managed in
the Notes panel — add one by selecting lines in the diff and switching the draft type
to Question (Ctrl+t cycles comment → question → note); AI replies appear there.
Validate / re-anchor findings
| App | Invocation |
|---|---|
| TUI | Press a → Validate review |
| Desktop | Press Cmd+K → AI → Validate / re-anchor |
Re-checks existing findings against the current diff and re-anchors line references that have moved. Useful after making changes in response to review feedback.
Generate guided tour
TUI: Press a → Generate guided tour; run it again to regenerate.
Once a tour exists, a Tour entry appears in the mode list to read it.
Desktop: Press Cmd+K → AI → Generate tour
(or Regenerate tour if one already exists). A "Re-run guide" button also appears in the
Guide tab when the tour is stale. Writes tour.json.
Generate diagrams
| App | Invocation |
|---|---|
| Desktop | Press Cmd+K → AI → Diagrams (opens the Context tab), then pick a preset or enter a custom prompt |
AI-generated Mermaid diagrams of the current diff: mental model, subsystems, flows, or a custom description.
Writes diagrams/*.json (presets overwrite in place; custom diagrams accumulate). Rendered in the
right panel's Context tab. See
Desktop Features → Context tab.
Arena (multi-reviewer consensus)
Desktop only — and not in the ⌘K AI menu. Reach it from Cmd+K →
PR → Open AI review arena, or from the Review panel Arena
card: use Run as Arena or Promote to Arena there.
TUI: Arena is not available in the TUI.
Runs multiple reviewer agents in adversarial rounds: experts propose findings, fresh same-lens skeptics try to refute them, an arbiter rules kept / dropped / merged. Higher-confidence results than a single-pass review. See Arena in the desktop features guide.
Publish to GitHub (not a Hub action)
Publishing review comments to a GitHub PR is not an AI Hub action — it is a separate operation in both apps:
- TUI: Open the Git hub (g) → Push comments to GitHub to push local comments to the PR — as one review or as individual comments; press G to pull comments from GitHub.
- Desktop: Open the Comments card → Push to GitHub, then choose Push as review or push individual comments. Comments are pulled automatically when you open a PR tab.
Per-item AI actions desktop
Beyond the Hub-level actions above, the desktop app attaches small AI actions to individual items:
- Elaborate — on a question or finding: ask the AI to expand on it and answer any question directly.
- Validate with AI — on a note or finding: check whether it still holds against the current diff.
- Promote to comment — turn a private question, note, or AI finding into a GitHub comment; the original keeps a "Promoted to" link. This is the deliberate one-way bridge from your private stream into the shared PR space.
See Comments & Questions for how questions, notes, and comments differ.
Review philosophy
Every reviewer shares one severity model and a "what not to flag" list, documented in
skills/REVIEW_PHILOSOPHY.md and skills/REVIEW_RULES.md:
- P0 (high) = must fix · P1 (medium) = should fix · P2 (low) = nice to fix.
- Never flag naming, formatting, style, or file moves.
The professor follows a separate skills/PROFESSOR_PHILOSOPHY.md — it teaches rather than critiques.
Multi-reviewer runs
You can run the general review, any experts, and the professor concurrently. er merges each fresh
sidecar (matching the current diff hash) into the in-memory review, and every finding carries an agent
pill — General, Security, Professor, and so on — so you can tell at a
glance which agent raised it. The desktop app makes this a first-class flow: use Run reviewers…
in the ⌘K AI menu to select and launch multiple reviewers at once, or use the
review arena for adversarial multi-model consensus.
Testing without running a review
To see the overlays without invoking an AI tool, generate fixtures with a matching diff hash:
cd your-repo
bash /path/to/easy-review/scripts/generate-test-fixtures.sh
ER_REPO_LOCAL=1 er # read sidecars from the repo's .er/ directory