superbrowser-sdk 2.4.0__tar.gz → 2.6.0__tar.gz

{superbrowser_sdk-2.4.0 → superbrowser_sdk-2.6.0}/CHANGELOG.md

@@ -5,6 +5,106 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.6.0] — 2026-06-26
+
+### Added — MCP switch_tab navigation tool (P3.0A)
+
+- `switch_tab` is now a navigation-tier MCP tool (default-allowed, not
+  action-budgeted). Delegates to the existing `SuperBrowser.switch_tab(tab_id)`
+  facade method. Use `list_tabs` to discover tab IDs.
+- Default advertised tools: 13 → 14; action mode: 19 → 20.
+- Diagnostics remain session-wide after switching tabs; per-tab diagnostics
+  are not supported yet.
+
+### Added — MCP inspect-output redaction (P2.3)
+
+Inspect-tier MCP tools now redact detected secrets at the MCP serialization
+boundary before returning content to the agent. Redaction is **on by default**
+(matching `SecurityConfig.redaction_enabled=True`) and can be disabled via
+`SB_MCP_REDACTION=0` (env) or `SecurityConfig(redaction_enabled=False)`
+(programmatic).
+
+**Tools redacted (9):** `extract_text`, `observe`, `current_url`, `list_tabs`,
+`get_console_messages`, `get_page_errors`, `get_network_errors`, `list_requests`,
+`get_request`.
+
+**What gets redacted:**
+- Text fields (text, message, stack, title, failure_text) — secret patterns
+  detected by `SecretRedactor` (API keys, JWTs, passwords, PEM keys, etc.) are
+  replaced with `[REDACTED:{type}:{hash6}]` markers.
+- URL fields — two-pass: `redact_context()` redacts sensitive query-parameter
+  values (`[REDACTED:query_param]`), then `SecretRedactor` pattern-scans for
+  secrets in non-sensitive keys/fragments.
+
+**What does NOT change:**
+- Output shape (same keys/structure; only string values may contain markers).
+- Python SDK consumers (`sb.extract()`, `sb.observe()`, `sb.diagnostics`) —
+  redaction applies only at the MCP boundary, not the SDK path.
+- `screenshot` (visual content), `browser_status` (no sensitive content).
+- DiagnosticsBuffer capture behavior, facade outputs.
+
+### Compatibility notes
+
+- This is a **behavior change** for MCP consumers: inspect-tier output that
+  previously contained raw secrets will now contain redaction markers. Set
+  `SB_MCP_REDACTION=0` to restore raw output.
+
+### Fixed — Release tooling
+
+- `scripts/verify_release_artifact.py` now supports sdist (`.tar.gz`) artifacts
+  via `tarfile`, in addition to wheels (`.whl`) via `zipfile`. Previously the
+  verifier was wheel-only and threw `BadZipFile` on any sdist (hit during both
+  the v2.4.0 and v2.5.0 releases). Shared checks (distribution name, version,
+  extras, isolated install, import, CLI, README/docs scan) run for both archive
+  types; wheel-only checks (entry points, module shadowing guard) are skipped
+  for sdists with a clear "skipped" message. Unsupported archive types now fail
+  with a clear error instead of an opaque `BadZipFile`.
+
+## [2.5.0] — 2026-06-25
+
+### Added — MCP Server P2: Diagnostics Pack
+
+An explainability layer for failed reads/rendering. Five new inspect-tier
+tools let an agent inspect browser/page/network evidence when
+`navigate → wait_for → extract_text` returns unexpected content — without
+expanding the action surface. No `--allow-actions`, no action budget, no audit
+entry, no side effects.
+
+- `get_console_messages` — buffered browser console messages, filterable by
+  level.
+- `get_page_errors` — buffered uncaught page errors with stack traces.
+- `get_network_errors` — requests that failed (status ≥ 400, no response, or
+  net error).
+- `list_requests` — all buffered request summaries; returns `request_id` for
+  `get_request` follow-up.
+- `get_request` — one request's metadata by `request_id`.
+
+New `DiagnosticsBuffer` (`super_browser.agent.diagnostics`): session-wide ring
+buffers for console / page-error / network events, wired into the facade
+lifecycle (`start()` + `_attach_page()`). Survives tab switches. Exposed as
+`SuperBrowser.diagnostics`. Reads are snapshots (non-destructive); bounded
+deques (default 500, via `config.event_buffer_size`).
+
+### Design constraints (P2)
+
+- Diagnostics entries carry a monotonic `seq`, `timestamp_ms`, and `page_url`.
+- Request records use a stable `request_id` (assigned by the buffer). A URL can
+  have multiple requests (redirects, retries), so retrieval is always by
+  `request_id`, never by URL.
+- **No response bodies** and **no raw header values** are returned. `get_request`
+  exposes `header_names` (keys) only. Console text, page errors, and URLs are
+  returned as-is, matching the existing inspect-tool posture.
+- `get_request` returns a structured `{ok: false, reason: "not_found"}` when the
+  request was never buffered or evicted; it never raises.
+
+### Compatibility notes
+
+- The default stdio server now advertises **13 tools** (was 8): the 5
+  diagnostics tools are inspect-tier and always advertised. Clients asserting
+  on the advertised count should update. This is additive — no tools were
+  removed and no refusal strings changed.
+- Action mode advertises **19 tools** (was 14).
+
 ## [2.4.0] — 2026-06-23
 
 ### Added — MCP Server P1: Navigation Workflows

{superbrowser_sdk-2.4.0 → superbrowser_sdk-2.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: superbrowser-sdk
-Version: 2.4.0
+Version: 2.6.0
 Summary: Anti-detection agent browser SDK — stealth, budget governance, security guardrails, and structured error recovery behind one async facade
 Project-URL: Homepage, https://github.com/Octo-Lex/Super-Browser
 Project-URL: Documentation, https://github.com/Octo-Lex/Super-Browser#readme

superbrowser_sdk-2.6.0/docs/design/inspect_redaction.md

@@ -0,0 +1,277 @@
+# Design Note: Shared Inspect-Output Redaction
+
+> **Status:** Design proposal for review. No implementation yet.
+> **Scope:** P2.3 — define the redaction boundary before more default-readable MCP surfaces are added.
+> **Date:** 2026-06-25
+
+## Problem
+
+The inspect-tier MCP tools return raw page/browser content to the agent with no
+redaction:
+
+- `extract_text` — returns page text as-is
+- `observe` — returns URL, title, element counts (text in serialized form)
+- `get_console_messages` — returns console output as-is
+- `get_page_errors` — returns error messages + stack traces as-is
+- `list_requests` / `get_request` — returns header **names** only (by P2 design),
+  but URLs, failure text, and console/error content are unredacted
+- `get_network_errors` — returns failed request URLs + failure text as-is
+
+This is consistent today (all inspect tools have the same "no redaction"
+posture), and it was the deliberate P2 decision: ship diagnostics unredacted for
+parity with `extract_text`, then unify redaction across all inspect tools as a
+follow-up. **This document is that follow-up.**
+
+The risk: console messages, page errors, and URLs routinely contain secrets
+(API keys in console logs, tokens in redirect URLs, credentials in error
+messages). An agent that reads these via MCP and passes them to an LLM leaks
+secrets into the model context.
+
+## Existing machinery (reuse, don't rebuild)
+
+The codebase already has a complete redaction layer:
+
+| Component | Location | What it does |
+|---|---|---|
+| `SecretRedactor` | `security/redactor.py:61` | Pattern-based redaction of secrets in text (API keys, tokens, passwords, JWTs, etc.) |
+| `redact_args()` | `security/action_redaction.py:55` | Key-name + value-pattern redaction on dicts |
+| `redact_result_dict()` | `security/action_redaction.py:108` | Redacts `ActionResult.to_dict()` output (`data` + `error` fields) |
+| `redact_context()` | `security/action_redaction.py:81` | Scrubs sensitive query params from URLs |
+| `_SENSITIVE_KEYS` | `security/action_redaction.py:12` | Frozenset of sensitive key names (password, token, secret, cookie, auth, etc.) |
+| `configure_redaction()` | `security/action_redaction.py:45` | Sets the default redactor from a `SecurityConfig` |
+| `SecurityManager.check_action()` | `security/manager.py:111` | Runs redaction on action params when `redaction_enabled=True` (default) |
+
+**Critical gap:** `redact_result_dict()` is wired into `ActionResult.to_dict()`
+(`results/types.py:235`) — the **Python SDK serialization path**. But the MCP
+server's `_serialize_action_result()` (`mcp_server.py:432`) manually builds a
+payload from `ok`/`data`/`error`/`meta` and does **not** call
+`ActionResult.to_dict()` or `redact_result_dict()`. So the same `ActionResult`
+is redacted when serialized via the Python SDK but **not** when returned via
+MCP. Additionally, the diagnostics handlers (`_tool_get_console_messages`, etc.)
+and `_tool_current_url` build response dicts directly, bypassing
+`ActionResult` and `_serialize_action_result()` entirely.
+
+## Design principles
+
+1. **Do not destroy utility by default.** Redaction should mask known secret
+   patterns, not collapse all text. A redacted console message should still
+   show the message structure with secrets replaced by the existing markers
+   (`[REDACTED:{secret_type}:{hash6}]` or `[REDACTED:query_param]`).
+
+2. **Policy-driven, not always-on.** Redaction should be controlled by the
+   `SecurityConfig` (which is already `redaction_enabled=True` by default). The
+   MCP server constructs a `SecurityManager` in `run_server()`; the same config
+   should govern inspect-output redaction. An operator who wants raw output
+   (e.g. for debugging) can set `redaction_enabled=False`.
+
+3. **Redact at the MCP serialization boundary, not inside the browser.** The
+   browser captures raw content; redaction happens when the MCP server
+   serializes the response for the agent. This keeps the SDK fast path
+   unredacted (Python SDK consumers get raw data) while the agent-facing MCP
+   boundary applies the policy.
+
+4. **Masking format is documented and stable.** The existing redaction
+   machinery emits two marker formats, both preserved as-is:
+   - `SecretRedactor`: `[REDACTED:{secret_type}:{hash6}]` — e.g.
+     `[REDACTED:openai_key:abc123]` (includes the secret type and a 6-char
+     SHA-256 prefix for deduplication).
+   - `redact_context()`: `[REDACTED:query_param]` — for sensitive query
+     parameter values.
+   Agents can detect that redaction occurred without seeing the secret.
+
+5. **No output-shape change.** Redaction replaces string values in place; it
+   does not add, remove, or rename keys. A redacted `get_console_messages`
+   response has the same `{ok, messages, count}` shape — only `messages[i].text`
+   values may be masked.
+
+## Scope: which tools, what content
+
+### In scope (redaction applied)
+
+| Tool | Content to redact | Mechanism |
+|---|---|---|
+| `extract_text` | extracted text body | `SecretRedactor.redact()` on the text string |
+| `observe` | title, URL | URL redaction (see below); `SecretRedactor.redact()` on title |
+| `current_url` | page URL | URL redaction (see below) |
+| `list_tabs` | tab URLs, tab titles | URL redaction on each tab URL; `SecretRedactor.redact()` on titles |
+| `get_console_messages` | `text` field per entry | `SecretRedactor.redact()` on each `text` |
+| `get_page_errors` | `message`, `stack` fields | `SecretRedactor.redact()` on each |
+| `get_network_errors` | `url`, `failure_text` | URL redaction on URL; `SecretRedactor.redact()` on failure_text |
+| `list_requests` | `url` per entry | URL redaction on each URL |
+| `get_request` | `url`, `failure_text` | URL redaction on URL; `SecretRedactor.redact()` on failure_text |
+
+**URL redaction** (two-pass, applied to all URL fields):
+1. `redact_context(url)` — redacts query parameters whose key names match
+   `_SENSITIVE_KEYS` (token, access_token, api_key, cookie, etc.).
+2. `SecretRedactor.redact(result_url)` — pattern-scans the remaining URL for
+   secret substrings in non-sensitive query keys, fragments, or URL-like
+   strings (e.g. `?code=sk-...`, `?next=https://...token...`).
+3. Preserve scheme, host, and path where possible; redact only secret
+   substrings and sensitive query values.
+
+This two-pass approach is necessary because `redact_context()` alone only
+catches secrets in *named* sensitive parameters. Secrets in innocuously-named
+query values (`?code=sk-...`) or embedded in redirect URLs would leak through
+without the `SecretRedactor` pattern scan.
+
+### Already redacted (no change)
+
+| Content | Current state |
+|---|---|
+| Request/response header values | Not exposed (P2: `header_names` keys only) |
+| Response bodies | Not exposed (P2: excluded) |
+| Navigate URL/params | Redacted by `SecurityManager.check_action()` (navigation-tier) |
+| Action tool arguments | Redacted by `SecurityManager.check_action()` (action-tier) |
+
+### Out of scope
+
+- **`browser_status`** — returns runtime status (running, backend name). No
+  sensitive page content expected.
+- **`screenshot`** — visual content; secret-detection in image data is a
+  different problem (OCR + pattern matching). Deferred.
+- **Redaction of element HTML/attributes in `observe`.** `observe` returns
+  counts and metadata, not raw HTML. If a future tool exposes HTML, it joins
+  this scope.
+- **Per-tool redaction policy overrides.** All inspect tools share one policy
+  (the `SecurityConfig`). Per-tool granular control is deferred.
+
+## Architecture
+
+### Option A (recommended): shared MCP redaction helper
+
+Add a single helper in `mcp_server.py` that applies redaction to any
+inspect-tier response dict before serialization:
+
+```python
+def _redact_inspect_output(payload: dict[str, Any]) -> dict[str, Any]:
+    """Apply redaction policy to inspect-tier tool output.
+
+    Redacts string values in known-sensitive fields (text, message, url, etc.)
+    using the server's SecurityConfig. No-op if redaction is disabled or no
+    SecurityManager is configured.
+    """
+```
+
+Each inspect handler calls `_redact_inspect_output(payload)` before returning.
+The helper uses the existing `SecretRedactor` + `redact_context()` machinery —
+no new redaction logic.
+
+**Why this over alternatives:** it's the narrowest change (one helper, called
+from each handler), it reuses all existing machinery, and it keeps redaction at
+the MCP boundary (principle #3).
+
+### Option B (not recommended): wire `redact_result_dict` into `_serialize_action_result`
+
+This would redact any `ActionResult` returned via MCP. But most inspect tools
+don't return an `ActionResult` (they build dicts directly), so it wouldn't cover
+`get_console_messages`, `list_requests`, etc. Incomplete.
+
+### Option C (not recommended): redact inside `DiagnosticsBuffer` / the facade
+
+This would redact at capture time, contaminating the buffer with masked values.
+An SDK consumer using `sb.diagnostics` directly would get redacted data even if
+they want raw. Violates principle #3.
+
+## Sensitive classes (what gets redacted)
+
+Reuse the existing `SecretRedactor` pattern set (`security/redactor.py`), which
+covers:
+
+- API keys (Anthropic, OpenAI, OpenRouter, GitHub, AWS, Google, Slack, Stripe)
+- JWTs
+- Passwords
+- PEM/private keys
+- Database URLs
+- Generic tokens
+
+Plus `_SENSITIVE_KEYS` for key-name matching (cookie, auth, session_id, etc.).
+
+### Not redacted (by default)
+
+- **Emails and phone numbers.** Not secrets in the same risk class as API keys.
+  Can be added as a configurable pattern layer if needed, but deferred to avoid
+  false positives that break utility.
+- **IP addresses.** Same reasoning — operational data, not credentials.
+- **Payment card numbers (PAN).** The `SecretRedactor` does not currently detect
+  these. Could be added as a custom pattern via `SecurityConfig.custom_secret_patterns`.
+
+## Compatibility
+
+- **Default behavior changes:** with `redaction_enabled=True` (the default),
+  inspect output now has secrets masked. This is a **behavior change** for
+  existing MCP consumers who relied on raw output. Documented in CHANGELOG.
+- **Output shape unchanged:** same keys, same structure. Only string values may
+  contain redaction markers (`[REDACTED:{secret_type}:{hash6}]` from
+  `SecretRedactor`, or `[REDACTED:query_param]` from `redact_context()`).
+- **Disabling:**
+  - **Current programmatic path:** pass
+    `SecurityManager(SecurityConfig(redaction_enabled=False))` into
+    `run_server(..., security_manager=...)`. Note that `run_server()` accepts a
+    `SecurityManager`, not a `SecurityConfig`; `_build_default_security_manager()`
+    constructs a default `SecurityConfig` internally using only domain
+    allow/block lists from env.
+  - **Proposed operator/CLI path:** add `SB_MCP_REDACTION=0|false|off` env
+    toggle in the implementation PR, parsed by `_build_default_security_manager()`
+    alongside the existing domain-list env vars. This gives operators a simple
+    opt-out without constructing a custom `SecurityManager`.
+- **Backward-compat for SDK consumers:** Python SDK users calling
+  `sb.extract()` / `sb.observe()` directly are unaffected — redaction applies
+  only at the MCP serialization boundary.
+
+## Test plan
+
+### Unit tests (shared helper)
+
+- `_redact_inspect_output` masks a known API key in a text field
+- `_redact_inspect_output` masks a token in a URL query param
+- `_redact_inspect_output` is a no-op when redaction is disabled
+- `_redact_inspect_output` is a no-op when no secret patterns are present
+- Output shape preserved (same keys, same structure)
+
+### Tool-specific fixture tests
+
+- `extract_text`: fixture with an API key in the text → redacted in response
+- `observe`: fixture with a secret in title → redacted; URL query param redacted
+- `current_url`: fixture/current page URL with a token query param → URL redacted
+- `list_tabs`: fixture/tab list with secret-bearing URL/title → URL/title redacted
+- `get_console_messages`: fixture with a token in console text → redacted
+- `get_page_errors`: fixture with a secret in error message → redacted
+- `get_network_errors`: fixture with failed request URL + failure_text secret → both redacted
+- `list_requests`: fixture with a secret in URL query param → URL redacted
+- `get_request`: fixture with a secret in failure_text → redacted
+
+### Negative tests
+
+- Non-secret text passes through unchanged (no false positives on normal prose)
+- Redaction disabled → raw output (existing behavior)
+
+## Implementation order (after design approval)
+
+```text
+1. Add _redact_inspect_output() helper in mcp_server.py
+2. Wire into the 9 inspect-tier handlers (extract_text, observe, current_url,
+   list_tabs, 5 diagnostics)
+3. Add SB_MCP_REDACTION env toggle (proposed in this design)
+4. Unit tests for the helper
+5. Fixture tests per tool
+6. CHANGELOG [Unreleased] — behavior change note
+7. Verify: real-browser e2e with a test page that logs a fake API key
+```
+
+## Resolved questions (from design review)
+
+1. **Should `observe`'s URL be redacted?** **Yes.** Apply the two-pass URL
+   redaction (query-param names + secret pattern scan) while preserving scheme,
+   host, and path. The same policy applies to `current_url`, `list_tabs`,
+   diagnostics request URLs, and any future inspect-tier URL fields.
+
+2. **Should redaction be on by default in the MCP server?** **Yes.** Default-on
+   matches the existing `SecurityConfig(redaction_enabled=True)` default and is
+   the safer boundary for agent-facing MCP output. A documented opt-out is
+   required (both programmatic and env-based), because this is a behavior
+   change for MCP consumers.
+
+3. **Should console stack traces be truncated?** **No.** Use pattern-only
+   redaction by default. Preserve paths, line numbers, and stack structure;
+   redact detected secret substrings only. Truncation would hurt diagnostic
+   utility and should be a separate optional limit, not the default.

{superbrowser_sdk-2.4.0 → superbrowser_sdk-2.6.0}/docs/mcp.md

@@ -5,9 +5,10 @@ Super Browser exposes its browser inspection and control surface over the
 agents (Claude, Cursor, etc.) can observe and interact with a page without
 scripting Python.
 
-The stdio server advertises **eight tools by default** (six inspect + two
-navigation) — enough to read a URL end-to-end. When action mode is enabled,
-it advertises **six additional action tools**; every action call is checked by
+The stdio server advertises **fourteen tools by default** (six inspect + five
+diagnostics + three navigation) — enough to read a URL end-to-end and explain
+why a read failed. When action mode is enabled, it advertises **six additional
+action tools**; every action call is checked by
 `MCPSessionPolicy` and `SecurityManager` before it can reach the browser.
 
 See [RFC #178](https://github.com/Octo-Lex/Super-Browser/issues/178) for the
@@ -30,12 +31,12 @@ python -m patchright install chromium
 Either of:
 
 ```bash
-superbrowser-mcp                       # default: 8 tools (inspect + navigation)
-superbrowser-mcp --allow-actions       # 14 tools (adds the action tier)
+superbrowser-mcp                       # default: 14 tools (inspect + diagnostics + navigation)
+superbrowser-mcp --allow-actions       # 20 tools (adds the action tier)
 python -m super_browser.mcp_server
 ```
 
-Both start a **stdio** server. The default server advertises 8 tools and
+Both start a **stdio** server. The default server advertises 14 tools and
 recognizes (but refuses) action-tool calls with a structured policy refusal.
 
 Action mode can also be enabled via the environment:
@@ -118,18 +119,36 @@ advertised; the third requires action mode; the fourth is not implemented.
 | `extract_text` | `query` (required), `selector` (optional) | Text content, optionally scoped to a CSS selector. | Yes |
 | `screenshot` | `full_page` (optional, default `false`) | Base64 PNG of the viewport or full page. | Yes |
 | `list_tabs` | none | Snapshot of open tabs. | Yes |
+| `get_console_messages` | `level` (optional), `limit` (optional, default 100) | Buffered browser console messages (snapshot, non-destructive). | Yes |
+| `get_page_errors` | `limit` (optional, default 100) | Buffered uncaught page errors with stack traces. | Yes |
+| `get_network_errors` | `url_filter` (optional), `limit` (optional, default 100) | Requests that failed (status ≥ 400, no response, or net error). | Yes |
+| `list_requests` | `url_filter` (optional), `resource_type` (optional), `limit` (optional, default 100) | All buffered request summaries; returns `request_id` for `get_request`. | Yes |
+| `get_request` | `request_id` (required) | One request's metadata (method, url, status, `header_names` — keys only, no values, no body). | Yes |
+
+The five diagnostics tools read from a session-wide ring buffer that captures
+console messages, page errors, and network requests via page-event listeners.
+Reads are **snapshots** (non-destructive): the buffer is not cleared on read.
+Diagnostics entries carry a monotonic `seq`, a `timestamp_ms`, and the
+`page_url` at the time of the event. Request records use a stable `request_id`
+(assigned by the buffer); a URL can have multiple requests, so always retrieve
+via `request_id` from `list_requests`, not by URL. **No response bodies and no
+raw header values** are returned — `get_request` exposes `header_names` (keys)
+only.
 
 ### Navigation tier (always advertised)
 
-Navigation mutates browser state (page acquisition) but is default-allowed
-because reading requires a page to read. It is always checked by the
-`SecurityManager` and is audited, but it does **not** consume the action
-budget and does **not** require action mode.
+Navigation mutates browser state (page/context acquisition) but is
+default-allowed because reading requires page/context acquisition. `navigate`
+is checked and audited at the MCP boundary when a SecurityManager is
+configured. `switch_tab` delegates to the facade, which applies its configured
+facade security policy. Navigation tools do **not** consume the action budget
+and do **not** require action mode.
 
 | Tool | Args | Behavior |
 |---|---|---|
 | `navigate` | `url` (required), `wait_until` (optional) | Go to a URL. URL is passed to `SecurityManager` for injection detection, secret redaction, and domain allow/block enforcement. |
 | `wait_for` | exactly one of `selector` / `text` / `url` / `load_state`; `timeout_ms` (optional, 100–60000, default 10000) | Wait for a page condition before the next read. |
+| `switch_tab` | `tab_id` (required, integer) | Switch the active browser tab by ID (from `list_tabs`). Changes the page the agent reads from. Diagnostics remain session-wide after switching; per-tab diagnostics are not supported yet. |
 
 `wait_for` accepts exactly one condition per call (deterministic single
 results; compound waits can be added later as an explicit AND mode).
@@ -142,6 +161,24 @@ navigate → wait_for → observe / extract_text / screenshot
 
 This is the default-mode loop an agent uses to read a page.
 
+#### Debugging workflow
+
+When a read returns unexpected or empty content, the diagnostics tools explain
+why — without expanding the action surface:
+
+```text
+navigate → wait_for → extract_text / observe
+        ↓ (unexpected / empty / errored)
+get_console_messages → get_page_errors → get_network_errors
+        ↓ (inspect the evidence)
+list_requests → get_request(request_id=...)   (drill into one request)
+        ↓
+adjust selector / report failure / retry
+```
+
+Diagnostics are inspect-tier: no `--allow-actions`, no action budget, no audit
+entry, no side effects.
+
 ### Action tier (advertised when action mode is enabled)
 
 | Tool | Args | Security level |
@@ -193,7 +230,7 @@ restricts the survivors.
 
 The default server behavior partitions the surface by risk:
 
-- **`list_tools()`** advertises only the Inspect + Navigation tiers (8 tools).
+- **`list_tools()`** advertises only the Inspect + Navigation tiers (14 tools).
 - **`call_tool()`** still recognizes action-tool names and returns a structured
   policy refusal (`refusal.reason = "actions are disabled"`), not an "Unknown
   tool" error.

{superbrowser_sdk-2.4.0 → superbrowser_sdk-2.6.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "superbrowser-sdk"
-version = "2.4.0"
+version = "2.6.0"
 description = "Anti-detection agent browser SDK — stealth, budget governance, security guardrails, and structured error recovery behind one async facade"
 readme = "README.md"
 license = { text = "Apache-2.0" }