pi-lean-dimension 0.2.0 → 0.2.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -8,7 +8,7 @@ search via SearXNG — all toggled from a single `/web` command. When the toggle
8
8
  is off, the tools are removed from the agent's context entirely, so web browsing
9
9
  doesn't consume tokens or attention on sessions that aren't doing web work.
10
10
  The same surfaces are user-extensible: author navigation guides that resurface
11
- by domain, or drop in a stealth browser backend like [Camoufox](https://github.com/nichochar/camoufox)
11
+ by domain, or drop in a stealth browser backend like [Camoufox](https://github.com/daijro/camoufox)
12
12
  when a site blocks the shipped Chromium/Firefox.
13
13
 
14
14
  ## Quick start (recommended)
@@ -100,7 +100,7 @@ When search is installed, two independent glyphs appear:
100
100
  Beyond the `/web` toggle, two surfaces are user-driven rather than hardcoded:
101
101
 
102
102
  - **Navigation guides** — `web-learn` saves site-specific playbooks that auto-match by domain and resurface in later sessions.
103
- - **Custom browser backends** — if a site blocks the shipped Chromium/Firefox, drop a `bridge.py` subclass into `~/.pi/agent/pi-lean-portal/user-backends/` and drive a patched engine like [Camoufox](https://github.com/nichochar/camoufox) yourself. A quirks schema declares how the engine diverges from base Playwright, and `launch` options flow from `settings.json` to the subprocess at runtime. This is user-authored, user-audited code that the extension never auto-downloads — and as far as we're aware, no other Pi web plugin lets you run a browser backend you wrote yourself. Most installs never need it; the [portal README](packages/pi-lean-portal/README.md#stealth--custom-browser-backends) and [`contributed/README.md`](packages/pi-lean-portal/contributed/README.md) cover the full flow when you do.
103
+ - **Custom browser backends** — if a site blocks the shipped Chromium/Firefox, drop a `bridge.py` subclass into `~/.pi/agent/pi-lean-portal/user-backends/` and drive a patched engine like [Camoufox](https://github.com/daijro/camoufox) yourself. A quirks schema declares how the engine diverges from base Playwright, and `launch` options flow from `settings.json` to the subprocess at runtime. This is user-authored, user-audited code that the extension never auto-downloads — and as far as we're aware, no other Pi web plugin lets you run a browser backend you wrote yourself. Most installs never need it; the [portal README](packages/pi-lean-portal/README.md#stealth--custom-browser-backends) and [`contributed/README.md`](packages/pi-lean-portal/contributed/README.md) cover the full flow when you do.
104
104
 
105
105
  ---
106
106
 
@@ -199,7 +199,7 @@ All tool calls dispatch through the router. Key responsibilities:
199
199
  - **Atomic writes + concurrency safety** (`storage-state.ts`): `saveStorageState()` writes to a temp file then renames atomically, preventing half-write races. Concurrent writers merge at the cookie level (`name+domain+path` key) and localStorage level (`origin+name` key), so two agents sharing a named profile don't clobber each other's data.
200
200
  - **Session profiles** (`profile="session"`) are scoped to one pi conversation, stored under `_session-<piSessionId>`. Default profile is now `"session"` (changed from `"none"`), so conversations persist state automatically.
201
201
  - **Named profiles** (`profile="shopping"`, `profile="work"`) are shared across conversations and agents.
202
- - **Conversation-scoped default profile** set via `/web profile set <name>`, survives `/reload`/`/resume`.
202
+ - **Conversation-scoped default profile** set via `/web profile <name>` (or `/web profile none` / `/web profile session`), survives `/reload`/`/resume`.
203
203
  - **Cookie operations** (`getCookies`, `addCookies`, `clearCookies`) delegate to the browser plugin's Playwright `context.cookies()` / `context.clearCookies()`.
204
204
 
205
205
  ## Known Constraints & Debt
@@ -6,9 +6,7 @@
6
6
  > domain. A `/web` toggle removes the tools from the agent's context when
7
7
  > switched off, so web browsing doesn't consume tokens on sessions that aren't
8
8
  > doing web work. If a site blocks the shipped browsers, drop in your own
9
- > backend (e.g. [Camoufox](https://github.com/nichochar/camoufox)) — as far as
10
- > we're aware, no other Pi web plugin lets you run a browser backend you wrote
11
- > yourself.
9
+ > backend (e.g. [Camoufox](https://github.com/daijro/camoufox)).
12
10
  >
13
11
  > Part of the [pi-lean-dimension](https://github.com/coreyryanhanson/pi-lean-dimension)
14
12
  > web-tools suite. For SearXNG search support, install
@@ -42,7 +40,7 @@ npx playwright install chromium firefox
42
40
 
43
41
  Once loaded, you'll see a notification like:
44
42
 
45
- > 🌐 Browser extension loaded (plugins: chromium, firefox)
43
+ > 🌐 Browser extension loaded (plugins: chromium, firefox). Try: web-fetch for static pages or browser-navigate for interactive browsing.
46
44
 
47
45
  The browser tools are **enabled by default**. You can:
48
46
 
@@ -70,7 +68,7 @@ Beyond the toggle, two surfaces are user-extensible rather than hardcoded:
70
68
  [Navigation Guides](#navigation-guides-web-guide--web-learn).
71
69
  - **Custom browser backends** — if a site blocks the shipped Chromium/Firefox,
72
70
  drop a `bridge.py` subclass into `~/.pi/agent/pi-lean-portal/user-backends/`
73
- and drive a patched engine like [Camoufox](https://github.com/nichochar/camoufox)
71
+ and drive a patched engine like [Camoufox](https://github.com/daijro/camoufox)
74
72
  yourself. The full flow lives in [Backend Architecture](#backend-architecture)
75
73
  and [`contributed/README.md`](./contributed/README.md).
76
74
 
@@ -122,7 +120,8 @@ which defaults to `true`).
122
120
  **Auto-captured screenshots:** `browser-navigate` and `browser-snapshot` automatically
123
121
  capture a screenshot to a temp file (`/tmp/pi-lean-portal/screenshot-<taskId>.jpg`).
124
122
  Use the `read` tool to visually inspect the page when the accessibility tree isn't enough.
125
- No viewport resizing occurs — screenshots are captured at the native 1280px width.
123
+ Screenshots capture the 1280×720 viewport (not full-page) — the same viewport the
124
+ browser uses for navigation.
126
125
 
127
126
  ### 1. `browser-navigate` — Visit a Page
128
127
 
@@ -201,7 +200,7 @@ Navigates back in browser history. Returns the previous page's snapshot.
201
200
  browser-press key="Enter"
202
201
  ```
203
202
 
204
- Useful keys: `Enter`, `Tab`, `Escape`, `ArrowDown`, `ArrowUp`, `/`.
203
+ Useful keys: `Enter`, `Tab`, `Escape`, `ArrowDown`, `ArrowUp`, `Backspace`.
205
204
 
206
205
  ### 8. `browser-console` — Read Console / Run JS
207
206
 
@@ -306,12 +305,13 @@ Shows everything about the browser runtime in one notification:
306
305
  🌐 Browser tools: ✅ on | 📖 Learn mode: ❌ off
307
306
  ────────────────────────────────────────
308
307
  Status: idle
309
- Plugins: chromium
308
+ Plugins: chromium, firefox, chromium-py (disabled), firefox-py (disabled)
309
+ Use web-fetch for stateless HTTP fetches.
310
310
  Active sessions: 1
311
311
  PW [chromium] https://example.com — Example Domain [profile: session]
312
- Profiles: 2 on disk
313
- 📋 session (2.1 KB) ← active
314
- shopping (0.3 KB)
312
+ Profiles: 1 on disk (named)
313
+ shopping (0.3 KB) ← active
314
+ Session profiles: 1 (manage with /web profile)
315
315
  ```
316
316
 
317
317
  Covers:
@@ -320,7 +320,9 @@ Covers:
320
320
  - **Backend health** — idle, busy, or error state
321
321
  - **All registered plugins** — enabled/disabled status
322
322
  - **Active sessions** — current URL, title, profile name per session
323
- - **Profiles on disk** — state size and which one is currently active
323
+ - **Profiles on disk** — named profiles with state size and which is
324
+ currently active; session profiles are collapsed into a single count line
325
+ (inspect individually with `/web profile list`)
324
326
 
325
327
  When `pi-lean-search` is also installed, the status bar shows two independent
326
328
  glyphs: `● idle` (browser state) and `● searxng` (search health/state).
@@ -614,8 +616,9 @@ Size threshold for profile state warnings (default: 10 MB):
614
616
  ### Working with `@e` Element References
615
617
 
616
618
  - `@e1`, `@e2`, etc. are assigned based on the accessibility tree order
617
- - After clicking or scrolling, **always take a fresh snapshot** — old `@e`
618
- refs become stale
619
+ - `browser-click`/`type`/`scroll` already return a fresh snapshot and cache
620
+ the full tree to disk — no separate `browser-snapshot` needed unless you
621
+ want the uncompacted tree (`full=true`) or a screenshot
619
622
  - `browser-inspect` is cheaper than `browser-snapshot full=true` for finding
620
623
  specific elements
621
624
 
@@ -623,8 +626,9 @@ Size threshold for profile state warnings (default: 10 MB):
623
626
 
624
627
  - Snapshots are automatically compacted to ~2500 characters
625
628
  - Very large pages (>8000 chars) preserve the top ~2000 chars
626
- - The **full tree is cached to disk** at `/tmp/pi-lean-portal/snapshot-*.txt` —
627
- you can use `read` on the cache file with offset/limit
629
+ - The **full tree is cached to disk** at `/tmp/pi-lean-portal/snapshot-*.txt`
630
+ when it would otherwise be truncated — use `read` on the cache file with
631
+ offset/limit to retrieve the complete tree
628
632
  - `browser-inspect text=true query="keyword"` finds specific content without
629
633
  loading the full tree
630
634
 
@@ -636,8 +640,14 @@ When a page triggers anti-automation:
636
640
  2. The **bot-detection guide** footer appears with strategies available via `web-guide`
637
641
  3. If very few elements are detected (<5), the navigation is treated as
638
642
  a hard failure — the agent won't try to interact with a challenge page
639
- 4. Try `web-fetch` on the same URL — it sometimes succeeds where the
640
- interactive browser doesn't
643
+ 4. Retry with a stealth backend — the `browser-navigate` `strategy`
644
+ parameter lists registered backend names; a stealth backend (e.g.
645
+ `strategy="camoufox"`) can pass challenges the default `chromium`/
646
+ `firefox` triggers. Only names listed in the `strategy` description
647
+ are valid — there is no `"stealth"` alias
648
+ 5. Try `web-fetch` on the same URL — it skips JS execution, so it can
649
+ retrieve raw HTML on pages that block the interactive browser via
650
+ client-side fingerprinting (it won't help against server-side WAFs)
641
651
 
642
652
  ### Guide Creation Discipline
643
653
 
@@ -92,6 +92,34 @@ describe("loadFullConfig().browser", () => {
92
92
  mockGlobalSettings({ unrelated: true });
93
93
  const config = loadFullConfig().browser;
94
94
  expect(config.defaultProfile).toBe("session");
95
+ expect(config.maxStorageStateSize).toBe(10 * 1024 * 1024);
96
+ });
97
+
98
+ // ── maxStorageStateSize ────────────────────────────────────
99
+
100
+ describe("maxStorageStateSize", () => {
101
+ it("accepts a positive number", () => {
102
+ mockGlobalSettings({ maxStorageStateSize: 5 * 1024 * 1024 });
103
+ expect(loadFullConfig().browser.maxStorageStateSize).toBe(
104
+ 5 * 1024 * 1024,
105
+ );
106
+ });
107
+
108
+ it("floors fractional values", () => {
109
+ mockGlobalSettings({ maxStorageStateSize: 5.9 * 1024 * 1024 });
110
+ expect(loadFullConfig().browser.maxStorageStateSize).toBe(
111
+ Math.floor(5.9 * 1024 * 1024),
112
+ );
113
+ });
114
+
115
+ it("rejects zero / negative / non-finite and falls back to default", () => {
116
+ for (const bad of [0, -1, Infinity, NaN, "big", null]) {
117
+ mockGlobalSettings({ maxStorageStateSize: bad });
118
+ expect(loadFullConfig().browser.maxStorageStateSize).toBe(
119
+ 10 * 1024 * 1024,
120
+ );
121
+ }
122
+ });
95
123
  });
96
124
 
97
125
  // ── defaultProfile ─────────────────────────────────────────