admin/CC-Switch
1
0
Fork 0
mirror of https://github.com/farion1231/cc-switch.git synced 2026-07-28 16:56:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix(Security): zip-slip on skill install, two credential leaks, and three panic paths (#5811)

Browse Source 
* fix(security): harden GrokBuild credential handling and Codex/Anthropic transforms against malformed input

Three robustness/security fixes in the upstream v3.18.0 code, each with a
regression test.

1) grok_config: remove the unconditional XAI_API_KEY fallback in
   extract_credentials(). Credentials now come only from an explicit inline
   api_key or the process env var named by env_key. Silently substituting a
   different account's key (when the declared env_key var is unset) could
   leak that key to whatever base_url the config points at.

2) deeplink/provider: merge_grokbuild_config no longer resolves env vars
   into a plaintext api_key on import. A deeplink is untrusted input;
   resolving+inlining would persist the victim's environment secret into the
   imported provider's config.toml and ship it to the link's declared
   base_url. env_key now stays an indirection (name), not a resolved secret.

3) proxy transforms: stop panicking on malformed upstream data.
   - transform_codex_anthropic::anthropic_sse_to_message_value: only store a
     `message`/`content_block` when it is an object; otherwise treat it as
     empty, so the later `["content"]`/`["text"]`/`["signature"]` index
     assignments can't panic on a scalar/array Value.
   - streaming_codex_anthropic::responses_sse_events_from_anthropic_message:
     bail out with a failed-event when the buffered body is a top-level JSON
     array/scalar (a gateway that ignores stream:true), instead of
     index-assigning into a non-object.
   - mcp/grokbuild sync: normalize a non-table `mcp_servers` before
     inserting, avoiding a toml_edit IndexMut panic on a user-edited
     config.toml.

Panic findings verified with a minimal repro (serde_json index-assignment on
a non-object Value aborts). cargo test --lib: 2183 passed / 0 failed.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(mcp): normalize a non-table `mcp_servers` in Codex config before insert

Same class of bug as the GrokBuild MCP fix in the previous commit, but in the
much more widely used Codex path.

`sync_single_server_to_codex` guarded only with `contains_key("mcp_servers")`,
so a user-edited `~/.codex/config.toml` where the key exists but is *not* a
table (`mcp_servers = "x"` / `[]` / `42`) skipped the rebuild and then hit
`doc["mcp_servers"][id] = …`, which panics in toml_edit's `IndexMut`
(`.expect("index not found")`). The panic happens inside a Tauri command and
unwinds across the FFI boundary; in the provider-switch flow it also fires
after the DB/live write already committed, leaving a half-applied switch.

Extract the normalize-then-insert step into `upsert_mcp_server_table()` so the
previously panic-prone logic is unit-testable without touching the real
`~/.codex/config.toml`, and normalize any non-table value to an empty table
before inserting.

Audited the sibling MCP writers while here: claude.rs / gemini.rs /
hermes.rs / opencode.rs do not have this pattern (their `contains_key` uses are
read-only checks), and the other index-assignments in codex.rs operate on
freshly built `Table::new()` values, which are safe.

Tests: 2 new regression tests (malformed non-table values normalize and insert;
an existing valid table keeps its entries). cargo test --lib 2185 passed / 0
failed; cargo clippy --lib -D warnings clean.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(security): reject path-traversal entries when extracting a skill repo archive

`SkillService::download_and_extract` built the output path from the raw ZIP
entry name (`file.name()`), stripped only the leading `<repo>-<branch>/`
component, and then `dest.join(relative_path)` → `fs::File::create`. A crafted
entry such as `repo-main/../../../evil.sh` therefore escaped the destination
directory (zip-slip): arbitrary file write outside the temp extraction dir.

The archive is third-party controlled: it is downloaded from
`https://github.com/<owner>/<name>/archive/refs/heads/<branch>.zip`, and a
skill repo (`owner`/`name`) can be added through an untrusted `ccswitch://`
deeplink (`deeplink/skill.rs`), so the attacker fully controls the archive
contents.

Fix: resolve each entry through `zip::read::ZipFile::enclosed_name()`, which
rejects `..` components and absolute paths, before stripping the archive's
root directory and joining onto `dest`. Unsafe entries are skipped with a
warning. This matches what the two sibling extractors in this codebase already
did correctly (`extract_local_zip`, `webdav_sync/archive.rs`) — this call site
was the one that had been missed.

The extraction loop is split out into `extract_repo_archive()` so the guard is
testable without network access.

Verified the test actually catches the bug: with the guard reverted to the old
`file.name()` behaviour the new test fails ("zip-slip entry must not escape
dest (temp root)"); with the guard in place it passes.

cargo test --lib: 2186 passed / 0 failed; cargo clippy --lib -D warnings clean.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(security): strip all credentials from the shared Gemini common-config snippet

`extract_gemini_common_config` skipped only two hardcoded keys
(`GOOGLE_GEMINI_BASE_URL`, `GEMINI_API_KEY`) and copied every other `env`
entry into the shared snippet. But `GOOGLE_API_KEY` is a first-class Gemini
credential — `provider.rs` resolves it via
`first_non_empty(env, &["GEMINI_API_KEY", "GOOGLE_API_KEY"])` — so it was
never stripped.

That snippet is not inert: `apply_common_config_to_settings` (live.rs)
deep-merges it into the `env` of *every other* Gemini provider that uses the
common config, and the snippet is auto-extracted on startup, on import, and on
switch. Net effect: account A's key gets written into provider B and sent to
B's `GOOGLE_GEMINI_BASE_URL`, which may be a third-party relay. Anything else
the user put in `env` (`GOOGLE_APPLICATION_CREDENTIALS`, a proxy
`*_AUTH_TOKEN`, …) leaked the same way.

Fix: also skip `Self::is_sensitive_config_key(key)`, reusing the pattern
matcher the Claude extractor already relies on for exactly this reason (its
comment notes a fixed enumeration "will always miss the next `*_API_KEY`").
`GOOGLE_API_KEY` matches its `_API_KEY` suffix rule. Shareable non-secret
config (e.g. `GEMINI_TIMEOUT_MS`) is preserved.

Verified the test catches the bug: run against the old two-key filter it fails
with "credential GOOGLE_API_KEY must not leak into the shared Gemini snippet";
with the fix it passes.

cargo test --lib: 2187 passed / 0 failed; cargo clippy --lib -D warnings clean.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(security): validate skill directory from backups and sync-imported DB rows

The skill install pipeline sanitizes the install directory name, but two
entry points bypassed it entirely:

- restore_from_backup joined metadata.skill.directory (raw meta.json
  content) into the SSOT dir with no validation, allowing a crafted
  backup to copy attacker-controlled files outside the skills directory
  and to persist the poisoned value into the database.
- Sync import (WebDAV/S3) loads the remote database dump verbatim, so a
  malicious or compromised sync snapshot could plant a skills row whose
  directory contains path traversal. Every later raw join then operated
  on attacker-controlled paths — uninstall/remove_from_app would
  remove_dir_all outside the managed dirs (arbitrary directory deletion),
  and sync_to_app_dir would write/symlink outside them.

Add require_valid_directory() (built on the existing
sanitize_install_name) and enforce it at restore_from_backup,
sync_to_app_dir, remove_from_app, and uninstall. Regression tests cover
all three sinks plus the metadata path; each was verified against the
unguarded code first (disabling the guard makes the restore/uninstall
tests complete the traversal write/delete, failing the assertions).

Also fixes a pre-existing cargo fmt drift in mcp/grokbuild.rs.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(security): reject path separators in sanitize_install_name on all platforms

The previous components()-based check was platform-dependent: on
Linux/macOS a backslash is not a path separator, so "a\b" parsed as a
single Normal component and was accepted as a valid install name — the
same value becomes a nested path when synced or restored onto Windows.
CI caught this on the Linux runner (the Windows runner passed).

Reject both '/' and '\' explicitly so the validation is
platform-independent; the components() check stays as the second layer
for dot segments, roots, and prefixes.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(security): close the remaining skill-install attack chain

The zip-slip guard added earlier only covered half the problem, and the
repo coordinates that decide *which* archive gets downloaded were never
validated at all. Together those two gaps let an attacker choose both the
bytes and where they land.

1) zip-slip, second layer. `enclosed_name()` only guarantees the entry does
   not escape the *archive's own* root, and it does not normalise the path —
   `..` survives verbatim in the returned value. Stripping `root_name`
   afterwards spends one level of that depth budget, so `repo-main/../evil`
   still escaped `dest`. On Windows it is worse: `root_name` comes from
   `split('/')` and may contain backslashes, which Windows treats as
   separators, so one `strip_prefix` can eat N components. Re-check the
   *actual* relative path for `ParentDir` right before `dest.join()`.
   Verified by reverting the guard: the single-`..` case escapes without it.
   (The existing test used two `..`, which `enclosed_name()` rejects on its
   own — it covered the half that was already guarded.)

2) Repo coordinates are now validated. `download_repo` formats owner/name/
   branch straight into
   `https://github.com/{owner}/{name}/archive/refs/heads/{branch}.zip`,
   and `deeplink/skill.rs` applied no validation whatsoever. URL parsing
   resolves dot segments, so a branch of
   `../../../releases/download/v1/evil` retargets the request at a *release
   asset* — arbitrary attacker-uploaded bytes. That is what made (1)
   reachable in practice: git itself will not let a tree contain `..`.
   Note the trigger is below "install": repos are enabled by default, so
   merely opening the Skills panel downloads and extracts.

   `validate_repo_ref` whitelists all three fields. Branch names legally
   contain `/` (`feature/x`), so the check is per segment rather than a
   blanket separator ban; an empty branch keeps its existing "use the
   default branch" sentinel meaning, same as `HEAD`. The main guard sits in
   `download_repo` — the single convergence point for all four call paths —
   because both `skill_repos` and `skills` can be overwritten wholesale by a
   sync snapshot, which no insert-time check can prevent. Entry-point checks
   (deeplink, `add_skill_repo`, agents lock) exist so bad values surface
   immediately instead of silently failing on every later panel open.
   `build_skill_doc_url` is covered too: it feeds `readme_url`, which the
   frontend hands to `openExternal`.

3) Regressions from the previous commits in this branch, both fixed:
   - `uninstall` ran the directory guard *before* `db.delete_skill`. That
     method has exactly one call site and is not exposed as a command, so a
     row with a dirty `directory` (pre-v3.11.0 installs, or anything
     `import_from_apps` still creates today — it has no sanitiser) became
     permanently undeletable from the UI. Now the guard skips the filesystem
     work but still deletes the row.
   - `sync_to_app` propagated with `?`, so one bad row aborted the entire
     app's skill sync — and that runs on provider switch. Now it warns and
     continues per entry.
   - `require_valid_directory` returned `sanitize_install_name`'s normalised
     value, which trims. A DB value of `" foo "` would then join as `"foo"`
     and miss the real directory. It now validates without rewriting.

4) Guard applied to the sinks the earlier commits missed: `migrate_storage`
   (rename + remove_dir_all), `update_skill` (delete + write attacker-named
   paths), `import_from_apps` (both a sink and the source of dirty values —
   `selection.directory` arrives raw over IPC), `resolve_uninstall_backup_source`
   (copies any directory into the backup area, which the UI then lists),
   `check_updates`, and `migrate_skills_to_ssot`.

5) Extraction limits. The archive bytes were fully attacker-controlled via
   (2), and this path had no ceiling of any kind, unlike
   `webdav_sync/archive.rs`. Added entry count, total extracted bytes, a
   symlink-target cap (zip 2.4.2's `make_reader` does not truncate at the
   declared `uncompressed_size`, so a symlink-flagged entry that inflates to
   gigabytes was read straight into memory), per-directory charging (a
   directory-only archive writes no content bytes but still consumes inodes),
   and a download-body cap (`response.bytes()` buffered the whole archive
   before any limit applied). Budget is charged on bytes actually read and
   written, never on sizes declared in the archive header.

   Symlink materialisation is charged to the same budget, and a target that
   contains the link itself (`dir/link -> ..`) is now rejected: the existing
   "must stay inside base" check passes for it, and the recursive copy then
   re-copies its own output every level until PATH_MAX.

6) Temp directories are now RAII. `download_repo` and `extract_local_zip`
   called `TempDir::keep()` immediately and relied on every exit path
   remembering `remove_dir_all`. Several did not — including
   `fetch_repo_skills`, the highest-frequency path of all. Returning the
   guard makes the leak unrepresentable at the call sites.

* fix(config): make removals and edits survive user-authored config shapes

The previous commits hardened the *write* side of the MCP tables against a
non-table `mcp_servers`. The read/delete side kept using `as_table_mut`,
which returns None for an inline table (`mcp_servers = { foo = {...} }` —
valid TOML). Removal then silently did nothing: the UI reported success, the
entry stayed in the file, and Codex loaded it again on next start. That is
worse than the panic it mirrors, because users usually reach for the toggle
precisely when an MCP server is misbehaving. Both Codex and GrokBuild now use
`as_table_like_mut` on both sides.

Normalisation is no longer silent either. Replacing a user's hand-written
`mcp_servers = "x"` with an empty table destroys data, so all four sites that
do it now log a warning first.

`update_codex_toml_field` had the same asymmetry with a different symptom:
when `model_providers` or `[model_providers.<id>]` was an inline table,
`as_table_mut` returned None and execution fell through to the "write a
top-level field" fallback. The user's `base_url` edit landed at the wrong
level, Codex never read it, and nothing reported a problem.

`opencode_config` parses `~/.config/opencode/opencode.json` with json5 into a
bare Value and never checked the root's shape. A user file of `[]` or a
scalar made `set_provider`, `set_mcp_server` and `add_plugin` all panic on
index-assignment, inside a Tauri command, unwinding across the FFI boundary.
(A top-level `null` is fine — serde_json promotes it.) Rejecting a non-object
root at the single read site fixes all three. Rejecting rather than rebuilding
is deliberate: the file also holds the user's own `model` / `theme` settings,
so a silent rebuild would delete them. Same treatment for the `provider` and
`mcp` sections, whose non-object forms made writes silently no-op.

* fix(proxy): recover a malformed Anthropic content_block as text

Sanitising a non-object `content_block` to `{}` stops the panic but creates a
quieter failure: the final Responses conversion matches on the block's `type`
and silently drops anything it does not recognise, so a garbled block header
turned into a `completed` response with empty output. The client saw the model
say nothing, with no signal that data had been discarded.

The replacement now carries `type: "text"`. The deltas that follow a malformed
header are usually well-formed, so this recovers the common case; a tool-use
block still yields nothing, exactly as before. A warning is logged when the
substitution happens.

The regression test now asserts through `anthropic_response_to_responses`
rather than on the intermediate value — asserting on the intermediate alone
passes while the client still receives an empty response.

* fix(grokbuild): decouple base_url from credential resolution, reject env_key-only links

`resolve_usage_credentials` called `extract_credentials(...).unwrap_or_default()`,
so a missing credential blanked the base_url too — even though it is written
right there in the config. Removing the `XAI_API_KEY` fallback in the earlier
commit widened that considerably: a GUI process on macOS does not inherit the
shell environment, so an `env_key` that resolves fine in a terminal resolves to
nothing here. The result was a Base URL shown in the UI that differed from the
one actually used, `{{baseUrl}}` expanding to empty in usage scripts (turning
requests relative), and native balance queries reporting "API key is empty"
while hiding the real cause. The two values are now resolved independently, via
the existing `grok_config::extract_base_url`, matching how the Codex arm of the
same function already works.

A deeplink whose only credential is `env_key` is now rejected by name. It was
already unimportable — `build_grokbuild_settings` has no `env_key` slot — but it
failed with the generic "API key is required", which reads like a malformed link
and invites the obvious "just carry the name over" fix. Carrying it over is
exactly what must not happen: the forwarder and the usage query both resolve
`env_key` at request time, so the victim's environment secret would still reach
the link's `base_url`. Same leak, merely deferred. The message says so, and the
test sets the probe variable so that restoring the resolution turns it red.

* fix(security): scrub credentials already leaked into the Gemini shared snippet

Fixing the extractor only prevents new contamination. A Gemini snippet is never
re-extracted once it exists — startup auto-extract and post-import extraction
both require `snippet.is_none()`, and the on-switch rewrite only covers Claude
and Codex — so existing users keep injecting the leaked key into live config,
the proxy upstream, and the new-provider form, where another account's key is
plainly visible.

Removing it from the snippet alone would make things worse. Merge and strip
cancel out by *value equality*: on switch-away, `remove_common_config_from_settings`
deletes the injected keys using the snippet as its only record of what to look
for. Once the key is gone from the snippet, the residue left in live config is
backfilled verbatim into the victim provider's `settings_config`, turning a
transient leak into a permanent one. So the cleanup covers all four locations at
once, and step order is itself a safety property: every fallible step runs before
the irreversible one, and a failure returns an error so the next start retries
from an intact state.

Deletion is by key *and* value, never by key name alone, so a provider's own
same-named key with a different value survives. The `~/.gemini/.env` edit
preserves layout rather than round-tripping through
`parse_env_file`/`serialize_env_file`: that pair drops comments, blank lines and
unparseable lines, collapses duplicate keys and re-sorts the file. Acceptable
when re-projecting everything, destructive for a targeted removal the user never
asked for — in testing it reduced a six-line fixture to one line, taking the
user's own key with it.

An audit record is written before any provider row changes, listing key names and
affected provider ids but *no values*: `settings` is not in `SYNC_SKIP_TABLES`,
so it is uploaded by WebDAV/S3 sync, and these are precisely the credentials that
must be destroyed. Keeping values would trade one deletion for a plaintext copy
that spreads across devices, has no UI to reach it, and never expires. It is
written only when absent, so a partially-applied earlier run cannot overwrite the
original pre-cleanup state.

Consequence, intentional: some Gemini providers will now report a missing API key
and need one entered. That key was never theirs. (The victim's own value was
already overwritten at merge time and cannot be recovered — worth noting in the
release notes, along with a recommendation to rotate the leaked key.) The snippet
row is deleted rather than set to `{}` when nothing shareable remains, and the
`cleared` flag is deliberately not set — either would disable auto-extract
permanently and prevent the user's legitimate shared config from ever coming back.

The frontend snippet validator is aligned with the backend matcher. It had the
same two hardcoded key names, so a hand-edited snippet could put `GOOGLE_API_KEY`
straight back with no error.

---------

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Jason <farion1231@gmail.com>
This commit is contained in:
zayoka
2026-07-28 16:34:11 +08:00
committed by GitHub
parent ccda04bfa6
commit ff3bc242cc
21 changed files with 2992 additions and 215 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src-tauri/src/codex_config.rs
+61 -5
View File
@@ -1908,25 +1908,53 @@ pub fn update_codex_toml_field(toml_str: &str, field: &str, value: &str) -> Resu
if let Some(provider_key) = model_provider {
// Ensure [model_providers] table exists
if doc.get("model_providers").is_none() {
//
// 用 as_table_like_mut 而非 as_table_mut：用户把配置写成 inline table
// （`model_providers = { foo = {...} }`，TOML 合法）时 as_table_mut
// 返回 None，会一路掉进下面的顶层 fallback——用户改的 base_url 被写到
// 了错误层级且毫无提示。
if doc
.get("model_providers")
.is_none_or(|item| item.as_table_like().is_none())
{
// 键存在但不是表（`model_providers = 42`）时，下面这行会把用户
// 手写的值替换掉。旧代码在这种形状下会掉进顶层 fallback 而不动
// 它，所以归一化必须留痕——与 mcp/codex.rs、mcp/grokbuild.rs、
// opencode_config.rs 的同款处理保持一致。
if doc
.get("model_providers")
.is_some_and(|item| !item.is_none())
{
log::warn!("config.toml 的 model_providers 不是表，已重置为空表");
}
doc["model_providers"] = toml_edit::table();
}
if let Some(model_providers) = doc["model_providers"].as_table_mut() {
if let Some(model_providers) = doc
.get_mut("model_providers")
.and_then(toml_edit::Item::as_table_like_mut)
{
// Ensure [model_providers.<provider_key>] table exists
if !model_providers.contains_key(&provider_key) {
model_providers[&provider_key] = toml_edit::table();
model_providers.insert(&provider_key, toml_edit::table());
}
if let Some(provider_table) = model_providers[&provider_key].as_table_mut() {
if let Some(provider_table) = model_providers
.get_mut(&provider_key)
.and_then(toml_edit::Item::as_table_like_mut)
{
if trimmed.is_empty() {
provider_table.remove(field);
} else {
provider_table[field] = toml_edit::value(trimmed);
provider_table.insert(field, toml_edit::value(trimmed));
}
return Ok(doc.to_string());
}
}
log::warn!(
"config.toml 的 [model_providers.{provider_key}] 结构异常，{field} 改写为顶层字段"
);
}
// Fallback: no model_provider or structure mismatch → top-level field
@@ -2585,6 +2613,34 @@ model = "gpt-4"
assert_eq!(base_url, "https://fallback.api/v1");
}
#[test]
fn base_url_writes_into_inline_table_provider_section() {
// inline table 是合法 TOML，但 as_table_mut() 对它返回 None。旧代码会因此
// 掉进「写顶层字段」的 fallback：用户改的 base_url 落在错误层级，
// Codex 读不到，且界面毫无提示。
let input = r#"model_provider = "any"
model_providers = { any = { name = "any", base_url = "https://old.api/v1", wire_api = "responses" } }
"#;
let result = update_codex_toml_field(input, "base_url", "https://new.api/v1").unwrap();
let parsed: toml::Value = toml::from_str(&result).unwrap();
assert_eq!(
parsed["model_providers"]["any"]["base_url"].as_str(),
Some("https://new.api/v1"),
"must update the provider section, not a top-level field"
);
assert!(
parsed.get("base_url").is_none(),
"must not leak a top-level base_url fallback"
);
assert_eq!(
parsed["model_providers"]["any"]["wire_api"].as_str(),
Some("responses"),
"sibling fields must survive"
);
}
#[test]
fn clearing_base_url_removes_only_from_correct_section() {
let input = r#"model_provider = "any"
src-tauri/src/commands/skill.rs
+4
View File
@@ -301,6 +301,10 @@ pub fn get_skill_repos(app_state: State<'_, AppState>) -> Result<Vec<SkillRepo>,
/// 添加技能仓库
#[tauri::command]
pub fn add_skill_repo(repo: SkillRepo, app_state: State<'_, AppState>) -> Result<bool, String> {
// 整个结构体由前端反序列化而来，owner/name/branch 会被拼进归档下载 URL。
// 主防线在 download_repo，这里让非法值当场报错而不是沉淀进表。
SkillService::validate_repo_ref(&repo.owner, &repo.name, &repo.branch)
.map_err(|e| e.to_string())?;
app_state
.db
.save_skill_repo(&repo)
src-tauri/src/deeplink/provider.rs
+134 -3
View File
@@ -832,9 +832,34 @@ fn merge_grokbuild_config(
.as_ref()
.is_none_or(|value| value.is_empty())
{
request.api_key = model.api_key.or_else(|| {
crate::grok_config::extract_credentials(&config_toml).map(|(_, api_key)| api_key)
});
// Only inline an explicitly-declared `api_key`. Do NOT resolve `env_key`
// (or any process env var) into a plaintext value here: a deeplink is
// untrusted input, and resolving+inlining would silently persist the
// victim's environment secret into the imported provider's config.toml
// and ship it to whatever `base_url` the link declares. `env_key` is an
// indirection that must stay a name, not a resolved secret, on import.
request.api_key = model.api_key;
// An `env_key`-only link is not importable at all, and saying so beats
// falling through to the generic "API key is required" (which reads like
// a malformed link and invites a "just carry the name over" fix).
// Carrying the name over is exactly what must not happen: the forwarder
// and the usage query both resolve `env_key` at request time, so the
// victim's environment secret would still reach the link's `base_url`
// — the same leak, merely deferred.
if request
.api_key
.as_ref()
.is_none_or(|value| value.is_empty())
&& model.env_key.is_some()
{
return Err(AppError::InvalidInput(
"This link supplies its API key indirectly through `env_key`, which cannot be \
imported from an untrusted link. Add the provider manually and enter the key \
yourself."
.to_string(),
));
}
}
if request
.endpoint
@@ -929,6 +954,112 @@ mod tests {
}
}
/// deeplink 同时声明 `api_key` 与 `env_key` 时，导入结果只保留用户可见的
/// `api_key`：既不能带上解析后的明文环境变量，也不能把 `env_key` 这个间接
/// 引用本身写进供应商配置。
///
/// 后者容易被误当成「应该补上的功能」，但恰恰不能补：deeplink 是不可信输入，
/// 攻击者可以让 `env_key` 指向 `XAI_API_KEY`、`base_url` 指向自己的服务器。
/// 密钥虽然导入时没落盘，却会在转发/用量查询调用 `extract_credentials` 时
/// 被现场解析并发给攻击者——等于把已修掉的泄露换成延迟触发的版本。
/// 手工建供应商的表单路径不受影响，那里的 env_key 是用户自己输入的。
#[test]
#[serial_test::serial]
fn grokbuild_deeplink_never_carries_env_key_or_resolved_secret() {
let original = std::env::var_os("CC_SWITCH_DEEPLINK_ENV_PROBE");
std::env::set_var("CC_SWITCH_DEEPLINK_ENV_PROBE", "secret-must-not-leak");
let mut request = DeepLinkImportRequest {
resource: "provider".to_string(),
app: Some("grokbuild".to_string()),
name: Some("Attacker".to_string()),
..Default::default()
};
let config = serde_json::json!({
"config": concat!(
"[models]\ndefault = \"grok-env\"\n\n",
"[model.\"grok-env\"]\nmodel = \"grok-4.5\"\n",
"base_url = \"https://attacker.example/v1\"\nname = \"Attacker\"\n",
"api_key = \"sk-declared-by-link\"\n",
"env_key = \"CC_SWITCH_DEEPLINK_ENV_PROBE\"\n",
"api_backend = \"responses\"\ncontext_window = 500000\n",
)
});
merge_grokbuild_config(&mut request, &config).expect("merge should succeed");
let settings = build_grokbuild_settings(&request);
let rendered = settings["config"].as_str().expect("config string");
assert!(
!rendered.contains("secret-must-not-leak"),
"the environment secret must never be inlined: {rendered}"
);
assert!(
!rendered.contains("env_key"),
"the env_key indirection must not be carried over from an untrusted link: {rendered}"
);
assert!(
rendered.contains("api_key = \"sk-declared-by-link\""),
"only the explicitly declared api_key should survive: {rendered}"
);
match original {
Some(value) => std::env::set_var("CC_SWITCH_DEEPLINK_ENV_PROBE", value),
None => std::env::remove_var("CC_SWITCH_DEEPLINK_ENV_PROBE"),
}
}
/// `env_key` 独苗的链接必须在**公开入口**上被明确拒绝。
///
/// 这条走 `parse_and_merge_config` 而不是内部 helper：真实失败路径在这里，
/// 而且报错必须指名 `env_key`——否则用户只看到泛化的 "API key is required"，
/// 读起来像链接坏了，下一步就会有人「顺手把 env_key 透传过去」，把泄露改成
/// 延迟触发的版本。
#[test]
#[serial_test::serial]
fn grokbuild_env_key_only_link_is_rejected_at_the_public_entry() {
use base64::prelude::*;
// 探针变量必须**真的设上**。若它在环境里不存在，旧代码的
// `extract_credentials` 回退也解析不出东西，api_key 同样留空、同样触发
// 拒绝——测试就退化成只覆盖"没有 key 时报错"，对"不得解析环境变量"这条
// 真正的安全属性零覆盖。设上之后，一旦有人恢复回退解析，api_key 会变成
// 非空、拒绝不再触发，这条断言立刻变红。
let original = std::env::var_os("CC_SWITCH_DEEPLINK_ENV_PROBE");
std::env::set_var("CC_SWITCH_DEEPLINK_ENV_PROBE", "secret-must-not-leak");
let config_toml = concat!(
"[models]\ndefault = \"grok-env\"\n\n",
"[model.\"grok-env\"]\nmodel = \"grok-4.5\"\n",
"base_url = \"https://attacker.example/v1\"\nname = \"Attacker\"\n",
"env_key = \"CC_SWITCH_DEEPLINK_ENV_PROBE\"\n",
"api_backend = \"responses\"\ncontext_window = 500000\n",
);
let request = DeepLinkImportRequest {
resource: "provider".to_string(),
app: Some("grokbuild".to_string()),
name: Some("Attacker".to_string()),
config: Some(BASE64_STANDARD.encode(config_toml)),
config_format: Some("toml".to_string()),
..Default::default()
};
let result = parse_and_merge_config(&request);
match original {
Some(value) => std::env::set_var("CC_SWITCH_DEEPLINK_ENV_PROBE", value),
None => std::env::remove_var("CC_SWITCH_DEEPLINK_ENV_PROBE"),
}
let err = result.expect_err(
"an env_key-only link must not be importable, and its env var must never be resolved",
);
assert!(
err.to_string().contains("env_key"),
"the rejection must name env_key so it is not mistaken for a malformed link: {err}"
);
}
#[test]
fn build_hermes_settings_emits_snake_case() {
let settings = build_hermes_settings(&hermes_request());
src-tauri/src/deeplink/skill.rs
+14 -1
View File
@@ -33,12 +33,25 @@ pub fn import_skill_from_deeplink(
}
let owner = parts[0].to_string();
let name = parts[1].to_string();
let branch = request.branch.unwrap_or_else(|| "main".to_string());
// deeplink 是不可信输入，且 branch 会被拼进归档下载 URL：URL 解析会消解点段，
// `../../../releases/download/v1/evil` 之类会把落点改写成攻击者可上传的
// release asset。主防线在 SkillService::download_repo，这里是入库前的纵深拦截，
// 让用户当场看到错误而不是让脏数据沉淀进 skill_repos 表。
crate::services::skill::SkillService::validate_repo_ref(&owner, &name, &branch).map_err(
|_| {
AppError::InvalidInput(format!(
"Invalid skill repository reference: '{owner}/{name}' branch '{branch}'"
))
},
)?;
// Create SkillRepo
let repo = SkillRepo {
owner: owner.clone(),
name: name.clone(),
branch: request.branch.unwrap_or_else(|| "main".to_string()),
branch,
enabled: request.enabled.unwrap_or(true),
};
src-tauri/src/gemini_config.rs
+64 -2
View File
@@ -152,8 +152,71 @@ pub fn read_gemini_env() -> Result<HashMap<String, String>, AppError> {
Ok(parse_env_file(&content))
}
/// 从 .env 原文中按「键名 + 值」双匹配删除若干行，其余内容逐字保留
///
/// 不走 `parse_env_file` → `serialize_env_file` 的往返：那对函数会丢掉注释、空行、
/// 无法识别的行和重复定义，并按键名重排整个文件。全量投影时这无所谓（本来就要重写
/// 整份），但用来做**定向**清理就等于顺手把用户手写的东西一起删了。
///
/// 按值匹配而非按键名：只清掉扩散出去的那一份，用户自己写的同名不同值的行保留。
/// 同一个键有重复定义时也只删命中的那条，被它遮住的上一条会重新生效——这正是想要的
/// 结果，因为遮住它的恰恰是泄漏值。
///
/// 返回 `None` 表示没有任何一行命中，调用方据此跳过写盘。
pub fn remove_env_entries_preserving_layout(
content: &str,
doomed: &HashMap<String, String>,
) -> Option<String> {
let mut removed = false;
let mut kept: Vec<&str> = Vec::new();
for line in content.split('\n') {
let trimmed = line.trim();
let hit = !trimmed.is_empty()
&& !trimmed.starts_with('#')
&& trimmed.split_once('=').is_some_and(|(key, value)| {
doomed
.get(key.trim())
.is_some_and(|doomed_value| doomed_value == value.trim())
});
if hit {
removed = true;
} else {
kept.push(line);
}
}
removed.then(|| kept.join("\n"))
}
/// 从 `~/.gemini/.env` 中定向删除「键=值」完全匹配的行，返回是否真的改了文件
pub fn remove_gemini_env_entries(doomed: &HashMap<String, String>) -> Result<bool, AppError> {
let path = get_gemini_env_path();
if !path.exists() {
return Ok(false);
}
let content = fs::read_to_string(&path).map_err(|e| AppError::io(&path, e))?;
match remove_env_entries_preserving_layout(&content, doomed) {
Some(cleaned) => {
write_gemini_env_text_atomic(&cleaned)?;
Ok(true)
}
None => Ok(false),
}
}
/// 写入 Gemini .env 文件（原子操作）
pub fn write_gemini_env_atomic(map: &HashMap<String, String>) -> Result<(), AppError> {
write_gemini_env_text_atomic(&serialize_env_file(map))
}
/// 写入 Gemini .env 文件（原子操作，内容逐字落盘）
///
/// 与 `write_gemini_env_atomic` 共用目录/文件权限处理，区别只在于内容不经
/// `serialize_env_file` 归一化——供保序的定向删除使用。
pub fn write_gemini_env_text_atomic(content: &str) -> Result<(), AppError> {
let path = get_gemini_env_path();
// 确保目录存在
@@ -172,8 +235,7 @@ pub fn write_gemini_env_atomic(map: &HashMap<String, String>) -> Result<(), AppE
}
}
let content = serialize_env_file(map);
write_text_file(&path, &content)?;
write_text_file(&path, content)?;
// 设置文件权限为 600（仅所有者可读写）
#[cfg(unix)]
src-tauri/src/grok_config.rs
+59 -16
View File
@@ -209,22 +209,23 @@ pub fn extract_model_config(config_toml: &str) -> Option<GrokModelConfig> {
pub fn extract_credentials(config_toml: &str) -> Option<(String, String)> {
let config = extract_model_config(config_toml)?;
let api_key = config
.api_key
.or_else(|| {
config
.env_key
.as_deref()
.and_then(|key| std::env::var(key).ok())
.map(|value| value.trim().to_string())
.filter(|value| !value.is_empty())
})
.or_else(|| {
std::env::var("XAI_API_KEY")
.ok()
.map(|value| value.trim().to_string())
.filter(|value| !value.is_empty())
})?;
// Credentials only come from two explicit, config-declared sources:
// 1. an inline `api_key`, or
// 2. the process env var named by `env_key`.
//
// Deliberately NO unconditional fallback to `XAI_API_KEY`: silently
// substituting a different account's key (when the declared `env_key` var is
// unset) would leak that key to whatever `base_url` this config points at.
// An unset/missing declared credential must surface as "no credential"
// (None) so callers can fail loudly rather than transmit the wrong secret.
let api_key = config.api_key.or_else(|| {
config
.env_key
.as_deref()
.and_then(|key| std::env::var(key).ok())
.map(|value| value.trim().to_string())
.filter(|value| !value.is_empty())
})?;
Some((config.base_url, api_key))
}
@@ -540,6 +541,48 @@ context_window = 500000
}
}
/// 构造一个 `env_key` 指向未设置环境变量的 config——这是"声明了间接引用但
/// 该变量不存在"的场景，修复前会静默兜底到 `XAI_API_KEY`。
fn env_key_unset_config() -> &'static str {
r#"[models]
default = "grok-env"
[model."grok-env"]
model = "grok-4.5"
base_url = "https://attacker.example/v1"
name = "Attacker Env"
env_key = "GROK_TEST_DEFINITELY_UNSET_VAR"
api_backend = "responses"
context_window = 500000
"#
}
#[test]
#[serial]
fn does_not_fall_back_to_xai_api_key_when_declared_env_key_is_unset() {
// 即使进程里恰好设了 XAI_API_KEY，也不能被静默借用到别的 base_url 上。
let original_xai = std::env::var_os("XAI_API_KEY");
let original_unset = std::env::var_os("GROK_TEST_DEFINITELY_UNSET_VAR");
std::env::set_var("XAI_API_KEY", "xai-secret-should-not-leak");
std::env::remove_var("GROK_TEST_DEFINITELY_UNSET_VAR");
let credentials = extract_credentials(env_key_unset_config());
assert!(
credentials.is_none(),
"declared env_key unset must yield None, never a borrowed XAI_API_KEY; got {credentials:?}"
);
match original_xai {
Some(value) => std::env::set_var("XAI_API_KEY", value),
None => std::env::remove_var("XAI_API_KEY"),
}
match original_unset {
Some(value) => std::env::set_var("GROK_TEST_DEFINITELY_UNSET_VAR", value),
None => std::env::remove_var("GROK_TEST_DEFINITELY_UNSET_VAR"),
}
}
#[test]
fn strips_projected_mcp_servers_without_touching_model_config() {
let mut settings = json!({
src-tauri/src/lib.rs
+11
View File
@@ -1184,6 +1184,17 @@ pub fn run() {
}
}
// 必须排在 auto-extract 之前：先把历史泄漏进 Gemini 共享片段的凭据
// 清干净，否则紧接着的提取会基于被污染的 live 再写一遍。
if let Err(e) =
crate::services::provider::ProviderService::scrub_leaked_gemini_common_config(
&state,
)
.await
{
log::warn!("清理 Gemini 通用配置泄漏凭据失败: {e}");
}
initialize_common_config_snippets(&state);
// 检查 settings 表中的代理状态，自动恢复代理服务
src-tauri/src/mcp/codex.rs
+153 -22
View File
@@ -348,6 +348,74 @@ pub fn sync_enabled_to_codex(config: &MultiAppConfig) -> Result<(), AppError> {
/// 将单个 MCP 服务器同步到 Codex live 配置
/// 始终使用 Codex 官方格式 [mcp_servers]，并清理可能存在的错误格式 [mcp.servers]
/// 把单个 MCP server 表写入 `[mcp_servers]`，并保证该键是「表」。
///
/// `~/.codex/config.toml` 是用户可手改的：若 `mcp_servers` 存在但不是表
/// （如 `mcp_servers = "x"` / `[]`），仅判 `contains_key` 会跳过重建，随后的
/// `doc["mcp_servers"][id] = …` 会触发 toml_edit 的 `IndexMut` panic
/// （panic 发生在 Tauri command 内、跨 FFI 展开）。这里统一归一化后再插入。
fn upsert_mcp_server_table(
doc: &mut toml_edit::DocumentMut,
id: &str,
table: toml_edit::Table,
) -> Result<(), AppError> {
if doc
.get_mut("mcp_servers")
.and_then(toml_edit::Item::as_table_like_mut)
.is_none()
{
// 键存在但不是表时，归一化会丢掉用户手写的那个值——必须留痕，
// 否则用户只会看到自己的改动凭空消失。
if doc.get("mcp_servers").is_some_and(|item| !item.is_none()) {
log::warn!("config.toml 的 mcp_servers 不是表，已重置为空表");
}
doc["mcp_servers"] = toml_edit::table();
}
let servers = doc
.get_mut("mcp_servers")
.and_then(toml_edit::Item::as_table_like_mut)
.ok_or_else(|| AppError::McpValidation("config.toml 的 mcp_servers 不是表".to_string()))?;
servers.insert(id, toml_edit::Item::Table(table));
Ok(())
}
/// 从 `[mcp_servers]`（以及历史错误格式 `[mcp.servers]`）中删除单个 MCP server。
///
/// 与 `upsert_mcp_server_table` 对称地使用 `as_table_like_mut`：用户若把配置写成
/// inline table（`mcp_servers = { foo = {...} }`，TOML 合法），`as_table_mut` 会返回
/// None 导致删除**静默失效**——界面提示已移除，条目却还在文件里，Codex 下次启动照样
/// 加载。这比 panic 更隐蔽，因为用户往往正是发现某个 MCP 有问题才来关它的。
///
/// 与写入分离成纯 doc 级函数，使守卫可脱离真实 `~/.codex/config.toml` 单测。
fn remove_mcp_server_from_doc(doc: &mut toml_edit::DocumentMut, id: &str) {
if let Some(item) = doc.get_mut("mcp_servers") {
// `Item::None` 是 toml_edit 的占位形态，不是用户写下的值——对它告警是噪音。
// 必须在取可变借用之前算出来。
let user_authored = !item.is_none();
match item.as_table_like_mut() {
Some(mcp_servers) => {
mcp_servers.remove(id);
}
None if user_authored => {
log::warn!("config.toml 的 mcp_servers 不是表，无法删除服务器 '{id}'");
}
None => {}
}
}
// 同时清理可能存在于错误位置的数据：[mcp.servers]（如果存在）
if let Some(mcp_table) = doc.get_mut("mcp").and_then(|t| t.as_table_like_mut()) {
if let Some(servers) = mcp_table
.get_mut("servers")
.and_then(|s| s.as_table_like_mut())
{
if servers.remove(id).is_some() {
log::warn!("从错误的 MCP 格式 [mcp.servers] 中清理了服务器 '{id}'");
}
}
}
}
pub fn sync_single_server_to_codex(
_config: &MultiAppConfig,
id: &str,
@@ -356,7 +424,6 @@ pub fn sync_single_server_to_codex(
if !should_sync_codex_mcp() {
return Ok(());
}
use toml_edit::Item;
// 读取现有的 config.toml
let config_path = crate::codex_config::get_codex_config_path();
@@ -383,16 +450,9 @@ pub fn sync_single_server_to_codex(
}
}
// 确保 [mcp_servers] 表存在
if !doc.contains_key("mcp_servers") {
doc["mcp_servers"] = toml_edit::table();
}
// 将 JSON 服务器规范转换为 TOML 表
let toml_table = json_server_to_toml_table(server_spec)?;
// 使用唯一正确的格式：[mcp_servers]
doc["mcp_servers"][id] = Item::Table(toml_table);
upsert_mcp_server_table(&mut doc, id, toml_table)?;
// 写回文件
let new_text = doc.to_string();
@@ -425,19 +485,7 @@ pub fn remove_server_from_codex(id: &str) -> Result<(), AppError> {
}
};
// 从正确的位置删除：[mcp_servers]
if let Some(mcp_servers) = doc.get_mut("mcp_servers").and_then(|s| s.as_table_mut()) {
mcp_servers.remove(id);
}
// 同时清理可能存在于错误位置的数据：[mcp.servers]（如果存在）
if let Some(mcp_table) = doc.get_mut("mcp").and_then(|t| t.as_table_mut()) {
if let Some(servers) = mcp_table.get_mut("servers").and_then(|s| s.as_table_mut()) {
if servers.remove(id).is_some() {
log::warn!("从错误的 MCP 格式 [mcp.servers] 中清理了服务器 '{id}'");
}
}
}
remove_mcp_server_from_doc(&mut doc, id);
// 写回文件
let new_text = doc.to_string();
@@ -683,6 +731,89 @@ pub(super) fn json_server_to_toml_table(spec: &Value) -> Result<toml_edit::Table
mod tests {
use super::*;
#[test]
fn upsert_normalizes_non_table_mcp_servers_without_panicking() {
// 用户手改过的 config.toml：mcp_servers 是字符串而不是表。
// 修复前 `doc["mcp_servers"][id] = …` 会 panic。
for malformed in [
"mcp_servers = \"x\"\n",
"mcp_servers = []\n",
"mcp_servers = 42\n",
] {
let mut doc = malformed
.parse::<toml_edit::DocumentMut>()
.expect("fixture parses");
let table = json_server_to_toml_table(&json!({
"type": "stdio",
"command": "npx"
}))
.expect("server table");
upsert_mcp_server_table(&mut doc, "echo", table)
.unwrap_or_else(|e| panic!("upsert must not fail for {malformed:?}: {e}"));
let servers = doc
.get("mcp_servers")
.and_then(|item| item.as_table_like())
.unwrap_or_else(|| panic!("mcp_servers must be normalized to a table"));
assert!(servers.contains_key("echo"));
}
}
#[test]
fn upsert_preserves_existing_servers_in_a_valid_table() {
let mut doc = "[mcp_servers.keep]\ncommand = \"keep\"\n"
.parse::<toml_edit::DocumentMut>()
.expect("fixture parses");
let table = json_server_to_toml_table(&json!({
"type": "stdio",
"command": "npx"
}))
.expect("server table");
upsert_mcp_server_table(&mut doc, "added", table).expect("upsert");
let servers = doc
.get("mcp_servers")
.and_then(|item| item.as_table_like())
.expect("table");
assert!(servers.contains_key("keep"), "existing server must survive");
assert!(servers.contains_key("added"));
}
#[test]
fn remove_deletes_from_inline_table_form_too() {
// inline table 是合法 TOML，但 as_table_mut() 对它返回 None——用它做守卫
// 会让删除静默失效：界面说移除成功，条目却还在，Codex 下次启动照样加载。
let mut doc = "mcp_servers = { drop = { command = \"x\" }, keep = { command = \"y\" } }\n"
.parse::<toml_edit::DocumentMut>()
.expect("fixture parses");
remove_mcp_server_from_doc(&mut doc, "drop");
let servers = doc
.get("mcp_servers")
.and_then(|item| item.as_table_like())
.expect("mcp_servers must still be table-like");
assert!(
!servers.contains_key("drop"),
"removal must work on the inline-table form"
);
assert!(servers.contains_key("keep"), "siblings must survive");
}
#[test]
fn remove_is_a_noop_on_non_table_mcp_servers() {
// 既不能 panic，也不能把用户手写的值悄悄抹掉
let mut doc = "mcp_servers = 42\n"
.parse::<toml_edit::DocumentMut>()
.expect("fixture parses");
remove_mcp_server_from_doc(&mut doc, "whatever");
assert_eq!(doc.to_string(), "mcp_servers = 42\n");
}
#[test]
fn http_headers_are_only_written_to_codex_http_headers() {
let table = json_server_to_toml_table(&json!({
src-tauri/src/mcp/grokbuild.rs
+37 -7
View File
@@ -148,10 +148,30 @@ pub fn sync_single_server_to_grokbuild(
AppError::McpValidation(format!("解析 Grok Build config.toml 失败: {e}"))
})?
};
if !doc.contains_key("mcp_servers") {
// 若 mcp_servers 缺失或存在但不是 table（如 `mcp_servers = "x"` / `[]`），
// 用空 table 归一化，避免后续 `doc["mcp_servers"][id] = …` 对非 table 索引
// 触发 toml_edit 的 IndexMut panic。用户手写的 config.toml 不可信。
// 判定走不可变的 `as_table_like`：借可变引用只为判空，会逼着下面再 get_mut 一次。
if doc
.get("mcp_servers")
.is_none_or(|item| item.as_table_like().is_none())
{
// 归一化会丢掉用户手写的那个非表值，必须留痕。
if doc.get("mcp_servers").is_some_and(|item| !item.is_none()) {
log::warn!("Grok Build config.toml 的 mcp_servers 不是表，已重置为空表");
}
doc["mcp_servers"] = toml_edit::table();
}
doc["mcp_servers"][id] = Item::Table(json_server_to_grokbuild_toml_table(server_spec)?);
let servers = doc
.get_mut("mcp_servers")
.and_then(toml_edit::Item::as_table_like_mut)
.ok_or_else(|| {
AppError::McpValidation("Grok Build config.toml 的 mcp_servers 不是表".to_string())
})?;
servers.insert(
id,
Item::Table(json_server_to_grokbuild_toml_table(server_spec)?),
);
crate::config::write_text_file(&path, &doc.to_string())
}
@@ -171,11 +191,21 @@ pub fn remove_server_from_grokbuild(id: &str) -> Result<(), AppError> {
return Ok(());
}
};
if let Some(servers) = doc
.get_mut("mcp_servers")
.and_then(toml_edit::Item::as_table_mut)
{
servers.remove(id);
// 与写入侧对称使用 as_table_like_mut：inline table 形态下 as_table_mut 返回
// None，删除会静默失效——界面显示已移除，Grok Build 下次启动仍会加载它。
if let Some(item) = doc.get_mut("mcp_servers") {
// `Item::None` 是 toml_edit 的占位形态，不是用户写下的值——对它告警是噪音。
// 必须在取可变借用之前算出来。
let user_authored = !item.is_none();
match item.as_table_like_mut() {
Some(servers) => {
servers.remove(id);
}
None if user_authored => {
log::warn!("Grok Build config.toml 的 mcp_servers 不是表，无法删除服务器 '{id}'");
}
None => {}
}
}
crate::config::write_text_file(&path, &doc.to_string())
}
src-tauri/src/opencode_config.rs
+106 -4
View File
@@ -95,12 +95,27 @@ pub fn read_opencode_config() -> Result<Value, AppError> {
}
let content = std::fs::read_to_string(&path).map_err(|e| AppError::io(&path, e))?;
json5::from_str(&content).map_err(|e| {
let value: Value = json5::from_str(&content).map_err(|e| {
AppError::Config(format!(
"Failed to parse OpenCode config: {}: {e}",
path.display()
))
})
})?;
// 根节点必须是对象：下游 set_provider / set_mcp_server / add_plugin 都对它做
// `config["key"] = …` 索引赋值，而 serde_json 只把 Null 自动升级成对象，
// 数组或标量会直接 panic（panic 发生在 Tauri command 内、跨 FFI 展开）。
//
// 这里选择报错而不是重建根节点：opencode.json 里还有 model / theme 等用户自有
// 配置，静默重建等于删掉它们。让用户自己修文件，与 read_claude_live 的做法一致。
if !value.is_object() {
return Err(AppError::Config(format!(
"OpenCode 配置文件根节点必须是 JSON 对象: {}",
path.display()
)));
}
Ok(value)
}
pub fn write_opencode_config(config: &Value) -> Result<(), AppError> {
@@ -123,7 +138,13 @@ pub fn get_providers() -> Result<Map<String, Value>, AppError> {
pub fn set_provider(id: &str, config: Value) -> Result<(), AppError> {
let mut full_config = read_opencode_config()?;
if full_config.get("provider").is_none() {
// 判空要连「存在但不是对象」一起算：否则下面 as_object_mut 拿不到，
// 写入会静默失效——界面显示添加成功而文件里没有。provider 段是 cc-switch
// 的投影区，归一化不会碰用户自有的 model / theme 等顶层配置。
if !full_config.get("provider").is_some_and(Value::is_object) {
if full_config.get("provider").is_some() {
log::warn!("opencode.json 的 provider 不是对象，已重置为空对象");
}
full_config["provider"] = json!({});
}
@@ -142,6 +163,8 @@ pub fn remove_provider(id: &str) -> Result<(), AppError> {
if let Some(providers) = config.get_mut("provider").and_then(|v| v.as_object_mut()) {
providers.remove(id);
} else if config.get("provider").is_some() {
log::warn!("opencode.json 的 provider 不是对象，无法删除供应商 '{id}'");
}
write_opencode_config(&config)
@@ -182,7 +205,10 @@ pub fn get_mcp_servers() -> Result<Map<String, Value>, AppError> {
pub fn set_mcp_server(id: &str, config: Value) -> Result<(), AppError> {
let mut full_config = read_opencode_config()?;
if full_config.get("mcp").is_none() {
if !full_config.get("mcp").is_some_and(Value::is_object) {
if full_config.get("mcp").is_some() {
log::warn!("opencode.json 的 mcp 不是对象，已重置为空对象");
}
full_config["mcp"] = json!({});
}
@@ -198,6 +224,8 @@ pub fn remove_mcp_server(id: &str) -> Result<(), AppError> {
if let Some(mcp) = config.get_mut("mcp").and_then(|v| v.as_object_mut()) {
mcp.remove(id);
} else if config.get("mcp").is_some() {
log::warn!("opencode.json 的 mcp 不是对象，无法删除服务器 '{id}'");
}
write_opencode_config(&config)
@@ -265,3 +293,77 @@ pub fn remove_plugins_by_prefixes(prefixes: &[&str]) -> Result<(), AppError> {
write_opencode_config(&config)
}
#[cfg(test)]
mod tests {
use super::*;
struct TestHomeGuard(Option<std::ffi::OsString>);
impl TestHomeGuard {
fn set(home: &std::path::Path) -> Self {
let guard = Self(std::env::var_os("CC_SWITCH_TEST_HOME"));
std::env::set_var("CC_SWITCH_TEST_HOME", home);
guard
}
}
impl Drop for TestHomeGuard {
fn drop(&mut self) {
match self.0.take() {
Some(value) => std::env::set_var("CC_SWITCH_TEST_HOME", value),
None => std::env::remove_var("CC_SWITCH_TEST_HOME"),
}
}
}
fn write_config(home: &std::path::Path, content: &str) {
let dir = home.join(".config").join("opencode");
std::fs::create_dir_all(&dir).expect("create config dir");
std::fs::write(dir.join("opencode.json"), content).expect("write config");
}
#[test]
#[serial_test::serial]
fn read_rejects_non_object_root_instead_of_panicking_downstream() {
let temp = tempfile::tempdir().expect("tempdir");
let _guard = TestHomeGuard::set(temp.path());
// 顶层数组/标量会让下游 `config["provider"] = …` 触发 serde_json panic。
// 顶层 null 例外——serde_json 会把它自动升级成对象，本来就不炸。
for malformed in ["[]", "[{\"a\":1}]", "42", "\"oops\""] {
write_config(temp.path(), malformed);
let result = read_opencode_config();
assert!(
result.is_err(),
"non-object root must be rejected: {malformed}"
);
}
write_config(temp.path(), "{\"model\": \"x\"}");
assert!(
read_opencode_config().is_ok(),
"a normal object config must still load"
);
}
#[test]
#[serial_test::serial]
fn set_mcp_server_normalizes_non_object_section() {
let temp = tempfile::tempdir().expect("tempdir");
let _guard = TestHomeGuard::set(temp.path());
// `"mcp": []` 时旧代码的 as_object_mut 返回 None → 写入静默失效
write_config(temp.path(), "{\"model\": \"keep-me\", \"mcp\": []}");
set_mcp_server("echo", json!({"command": "npx"})).expect("set must succeed");
let config = read_opencode_config().expect("reload");
assert_eq!(
config["mcp"]["echo"]["command"], "npx",
"server must actually be written"
);
assert_eq!(
config["model"], "keep-me",
"unrelated user config must be preserved"
);
}
}
src-tauri/src/provider.rs
+17 -5
View File
@@ -169,11 +169,23 @@ impl Provider {
let api_key = first_non_empty(env, &["GEMINI_API_KEY", "GOOGLE_API_KEY"]);
(base_url, api_key)
}
AppType::GrokBuild => settings
.get("config")
.and_then(Value::as_str)
.and_then(crate::grok_config::extract_credentials)
.unwrap_or_default(),
// GrokBuild 的 base_url 与 api_key 必须各自解析：extract_credentials 在
// 凭据缺失时整个 Option 变 None，一并 unwrap_or_default 会把明明写在
// 配置里的 base_url 也清成空串。凭据缺失是常态（env_key 指向的变量在
// GUI 进程里读不到），端点不该被连坐——否则用量脚本的 {{baseUrl}} 变成
// 相对路径、余额查询只报「API key is empty」掩盖真因。
// 与上面 Codex 分支的写法保持一致。
AppType::GrokBuild => {
let config_text = settings.get("config").and_then(Value::as_str);
let base_url = config_text
.and_then(crate::grok_config::extract_base_url)
.unwrap_or_default();
let api_key = config_text
.and_then(crate::grok_config::extract_credentials)
.map(|(_, api_key)| api_key)
.unwrap_or_default();
(base_url, api_key)
}
// Hermes (config.yaml) flattens credentials at the top level, snake_case.
AppType::Hermes => (
str_at(settings.get("base_url")),
src-tauri/src/proxy/providers/streaming_codex_anthropic.rs
+24
View File
@@ -589,6 +589,21 @@ pub(crate) fn responses_sse_events_from_anthropic_message(
tool_context: CodexToolContext,
) -> Vec<Bytes> {
let mut state = AnthropicToResponsesState::with_tool_context(tool_context);
// The whole conversion below assumes `body` is an Anthropic message object.
// A misbehaving gateway can return a top-level JSON array (or scalar) with
// HTTP 200 despite `stream:true`; index-assigning `message_start["content"]`
// on such a non-object would panic. Bail out gracefully instead.
if !body.is_object() {
return state
.failed_event(
"upstream returned a non-object Anthropic message body".to_string(),
Some("invalid_response".to_string()),
)
.into_iter()
.collect();
}
if body.get("type").and_then(Value::as_str) == Some("error") || body.get("error").is_some() {
let (message, error_type) = extract_anthropic_sse_error(body);
return state
@@ -819,6 +834,15 @@ mod tests {
assert!(!merged.contains("stream_truncated"));
}
#[test]
fn test_json_non_object_body_returns_failed_event_not_panic() {
// A gateway that ignores `stream:true` and returns a top-level JSON array
// would have panicked on `message_start["content"] = …` before the guard.
let merged = render_message_events(&json!([1, 2, 3]));
assert!(merged.contains("event: response.failed"));
assert!(merged.contains("invalid_response"));
}
#[test]
fn test_json_message_becomes_complete_responses_stream() {
let merged = render_message_events(&json!({
src-tauri/src/proxy/providers/transform_codex_anthropic.rs
+66 -2
View File
@@ -1418,13 +1418,39 @@ pub fn anthropic_sse_to_message_value(body: &str) -> Result<Value, ProxyError> {
};
match value.get("type").and_then(|t| t.as_str()).unwrap_or("") {
"message_start" => {
if let Some(msg) = value.get("message") {
// Only accept an object message; a malformed upstream could send a
// scalar/array here, and the later `message["content"] = …` index
// assignment would panic on a non-object Value.
if let Some(msg) = value.get("message").filter(|m| m.is_object()) {
*message = Some(msg.clone());
}
}
"content_block_start" => {
if let Some(index) = value.get("index").and_then(|v| v.as_u64()) {
let block = value.get("content_block").cloned().unwrap_or(json!({}));
// Sanitize to an object: any later index-assignment (`["text"]`,
// `["signature"]`, `["input"]`) requires a JSON object, so a
// malformed non-object block from the upstream cannot be stored
// verbatim (it would panic on the next delta).
//
// The replacement carries `type: "text"` rather than being empty:
// the deltas that follow are usually well-formed, and a block with
// no `type` is silently dropped by the final Responses conversion,
// which turns a garbled block header into a `completed` response
// with empty output — the client sees the model saying nothing and
// has no way to tell that data was discarded. A text block recovers
// the common case; a tool-use block still yields nothing, exactly as
// it did before.
let block = match value.get("content_block") {
Some(block) if block.is_object() => block.clone(),
malformed => {
if malformed.is_some() {
log::warn!(
"Anthropic upstream sent a non-object content_block at index {index}; recovering it as a text block"
);
}
json!({ "type": "text" })
}
};
blocks.insert(index, block);
json_accum.entry(index).or_default();
}
@@ -2953,4 +2979,42 @@ data: {\"type\":\"content_block_start\",\"index\":0,\"content_block\":{\"type\":
"data: {\"type\":\"message_start\",\"message\":{\"id\":\"m\",\"content\":[]}}\n\n";
assert!(anthropic_sse_to_message_value(sse).is_err());
}
#[test]
fn test_anthropic_sse_aggregation_non_object_content_block_does_not_panic() {
// A malformed upstream can send a non-object `content_block`; the index
// assignment on the next delta would have panicked before the shape guard.
let sse = concat!(
"data: {\"type\":\"message_start\",\"message\":{\"id\":\"m\",\"content\":[]}}\n\n",
"data: {\"type\":\"content_block_start\",\"index\":0,\"content_block\":[1]}\n\n",
"data: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"text_delta\",\"text\":\"x\"}}\n\n",
"data: {\"type\":\"message_stop\"}\n\n",
);
let msg = anthropic_sse_to_message_value(sse)
.expect("aggregation must not panic on a non-object content_block");
assert_eq!(msg["content"][0]["text"], json!("x"));
// Not panicking is only half of it: the sanitized block must still carry a
// `type`, because the final conversion matches on it and silently drops
// anything it does not recognise. Asserting only on the intermediate value
// would pass while the client receives a `completed` response with empty
// output and no indication that the text was thrown away.
let response = anthropic_response_to_responses(msg).expect("final conversion must succeed");
assert_eq!(
response["output"][0]["content"][0]["text"],
json!("x"),
"text recovered from a malformed block must survive to the Responses output: {response}"
);
}
#[test]
fn test_anthropic_sse_aggregation_non_object_message_errors_not_panic() {
// A malformed upstream can send a scalar `message`; the later
// `message["content"] = …` would have panicked before the shape guard.
let sse = concat!(
"data: {\"type\":\"message_start\",\"message\":\"oops\"}\n\n",
"data: {\"type\":\"message_stop\"}\n\n",
);
assert!(anthropic_sse_to_message_value(sse).is_err());
}
}
src-tauri/src/services/provider/mod.rs
+708 -5
View File
@@ -702,6 +702,469 @@ mod tests {
);
}
#[test]
fn extract_gemini_common_config_strips_credentials_keeps_shareable() {
// Gemini 的共享片段会被 deep-merge 回**其它** Gemini 供应商的 env
// (live.rs::apply_common_config_to_settings)，因此任何凭据都不得进入片段。
// 之前这里只硬编码跳过 GEMINI_API_KEY/GOOGLE_GEMINI_BASE_URL，而
// GOOGLE_API_KEY 是 provider.rs 认可的一等 Gemini 凭据 → 会泄露到别的供应商。
let settings = json!({
"env": {
"GEMINI_API_KEY": "g-gem",
"GOOGLE_API_KEY": "g-legacy-real-key",
"GOOGLE_GEMINI_BASE_URL": "https://gemini.example",
"GOOGLE_APPLICATION_CREDENTIALS": "/path/creds.json",
"SOME_PROXY_AUTH_TOKEN": "tok-proxy",
// 可共享的非机密配置必须保留
"GEMINI_TIMEOUT_MS": "30000"
}
});
let snippet =
ProviderService::extract_gemini_common_config(&settings).expect("extract should work");
let value: Value = serde_json::from_str(&snippet).expect("snippet is valid JSON");
for leaked in [
"GEMINI_API_KEY",
"GOOGLE_API_KEY",
"GOOGLE_APPLICATION_CREDENTIALS",
"SOME_PROXY_AUTH_TOKEN",
] {
assert!(
value.get(leaked).is_none(),
"credential {leaked} must not leak into the shared Gemini snippet"
);
}
assert_eq!(
value.get("GEMINI_TIMEOUT_MS").and_then(|v| v.as_str()),
Some("30000"),
"shareable non-secret config must be preserved"
);
}
/// 造一个「已被污染」的现场：片段里带 A 账号的凭据 + 一个合法可共享键。
#[test]
fn sensitive_key_matcher_covers_common_credential_namings() {
for key in [
// 裸 `_KEY`：最常见的写法，却曾被"只枚举 `_API_KEY` 这些子类"漏在外面
"OPENAI_KEY",
"GROQ_KEY",
"XAI_KEY",
// 不带分隔符的复合写法
"VOLC_ACCESSKEY",
"ALIYUN_SECRETKEY",
"SOME_APITOKEN",
// personal access token：既不含 TOKEN 也不含 KEY
"GITHUB_PAT",
"gitlab_pat",
// 口令类缩写
"MYSQL_PWD",
"DB_PASS",
"GPG_PASSPHRASE",
"AWS_CREDS",
] {
assert!(
ProviderService::is_sensitive_config_key(key),
"{key} must be treated as a credential"
);
}
// 后缀必须带下划线，不能把正常配置一起卷进来
for key in [
"PATH",
"OLDPWD",
"GEMINI_COMPAT",
"SSL_BYPASS",
"GEMINI_TIMEOUT_MS",
"CLAUDE_CODE_MAX_OUTPUT_TOKENS",
] {
assert!(
!ProviderService::is_sensitive_config_key(key),
"{key} is ordinary shareable config and must not be stripped"
);
}
}
fn seed_leaked_gemini_state(db: &Arc<Database>) {
db.set_config_snippet(
"gemini",
Some(
json!({
"GOOGLE_API_KEY": "key-A-leaked",
"SOME_PROXY_AUTH_TOKEN": "tok-A-leaked",
"GEMINI_TIMEOUT_MS": "30000"
})
.to_string(),
),
)
.expect("seed snippet");
// 受害者 B：泄漏的密钥已经被合并进它的 env
let victim = Provider::with_id(
"b".into(),
"Relay B".into(),
json!({ "env": {
"GOOGLE_GEMINI_BASE_URL": "https://relay-b.example",
"GOOGLE_API_KEY": "key-A-leaked",
"GEMINI_TIMEOUT_MS": "30000"
}}),
None,
);
db.save_provider("gemini", &victim).expect("save victim");
// 供应商 C：自己写了同名键但值不同，不能被误删
let unrelated = Provider::with_id(
"c".into(),
"Own Key C".into(),
json!({ "env": {
"GOOGLE_GEMINI_BASE_URL": "https://c.example",
"GOOGLE_API_KEY": "key-C-owned"
}}),
None,
);
db.save_provider("gemini", &unrelated).expect("save c");
}
#[tokio::test]
#[serial]
async fn scrub_gemini_removes_leaked_credentials_from_snippet_and_providers() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("scrub must succeed");
// 片段：凭据清掉，可共享配置保留
let snippet = db
.get_config_snippet("gemini")
.expect("read snippet")
.expect("snippet must still exist");
let snippet: Value = serde_json::from_str(&snippet).expect("valid json");
assert!(snippet.get("GOOGLE_API_KEY").is_none());
assert!(snippet.get("SOME_PROXY_AUTH_TOKEN").is_none());
assert_eq!(
snippet.get("GEMINI_TIMEOUT_MS").and_then(Value::as_str),
Some("30000"),
"shareable config must survive the scrub"
);
// 受害者 B：扩散过去的那一份被清掉
let providers = db.get_all_providers("gemini").expect("providers");
let victim_env = &providers["b"].settings_config["env"];
assert!(
victim_env.get("GOOGLE_API_KEY").is_none(),
"leaked key must be removed from the victim provider"
);
assert_eq!(
victim_env.get("GEMINI_TIMEOUT_MS").and_then(Value::as_str),
Some("30000"),
"non-credential config must not be touched"
);
}
#[tokio::test]
#[serial]
async fn scrub_gemini_keeps_a_providers_own_differently_valued_key() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("scrub must succeed");
// 这条最容易写错成「按键名一刀切」：C 自己的密钥值与片段不同，是它自己的凭据
let providers = db.get_all_providers("gemini").expect("providers");
assert_eq!(
providers["c"].settings_config["env"]
.get("GOOGLE_API_KEY")
.and_then(Value::as_str),
Some("key-C-owned"),
"a provider's own key must not be deleted by name matching"
);
}
#[tokio::test]
#[serial]
async fn scrub_gemini_audit_records_key_names_but_never_values() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("scrub must succeed");
let audit_text = db
.get_setting("gemini_common_config_scrub_audit_v1")
.expect("read audit")
.expect("an audit record must exist so the deletion is not silent");
// 值绝不能进这条记录：`settings` 会随 WebDAV/S3 同步上传，留值等于把一次
// 清除换成一份跨设备扩散、没有界面入口、永不过期的明文副本。
assert!(
!audit_text.contains("key-A-leaked") && !audit_text.contains("tok-A-leaked"),
"the audit record must never carry credential values: {audit_text}"
);
// 但必须说清楚删了什么、从哪删的，否则用户只能靠翻日志
let audit: Value = serde_json::from_str(&audit_text).expect("audit is JSON");
let removed: Vec<&str> = audit["removedFromSnippet"]
.as_array()
.expect("removedFromSnippet array")
.iter()
.filter_map(Value::as_str)
.collect();
assert!(
removed.contains(&"GOOGLE_API_KEY") && removed.contains(&"SOME_PROXY_AUTH_TOKEN"),
"every key removed from the snippet must be named: {audit}"
);
let victim = audit["providers"]
.as_array()
.expect("providers array")
.iter()
.find(|entry| entry["id"] == json!("b"))
.expect("every provider whose config gets rewritten must be recorded");
assert_eq!(
victim["removedKeys"],
json!(["GOOGLE_API_KEY"]),
"the record must name what was taken from each provider: {audit}"
);
}
#[tokio::test]
#[serial]
async fn scrub_gemini_never_overwrites_an_existing_audit_record() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
// 上一轮改到一半就中止的情形：完成标记没置位，下次启动会重跑，但那时
// 读到的"原始状态"已经残缺。无条件覆盖会拿残缺记录盖掉第一轮那份完整的。
db.set_setting(
"gemini_common_config_scrub_audit_v1",
"{\"from\":\"an earlier, complete run\"}",
)
.expect("seed an existing audit record");
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("scrub must succeed");
assert_eq!(
db.get_setting("gemini_common_config_scrub_audit_v1")
.expect("read audit")
.as_deref(),
Some("{\"from\":\"an earlier, complete run\"}"),
"an audit record from an earlier run must survive a retry"
);
}
#[tokio::test]
#[serial]
async fn scrub_gemini_cleans_the_live_env_without_a_current_provider() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
// 没有当前供应商——这正是 sync_current_provider_for_app 直接返回 Ok 而
// 根本不写文件的分支。此时 live 若清不掉，片段又已被清空，下次切换的
// backfill 就会把残留永久写进受害供应商的配置。
crate::gemini_config::write_gemini_env_atomic(&HashMap::from([
("GOOGLE_API_KEY".to_string(), "key-A-leaked".to_string()),
("GEMINI_TIMEOUT_MS".to_string(), "30000".to_string()),
// 只存在于 live 的手工修改：定向删除必须保住它，全量重投影会抹掉
(
"HTTPS_PROXY".to_string(),
"http://127.0.0.1:7890".to_string(),
),
]))
.expect("seed live env");
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("scrub must succeed");
let live = crate::gemini_config::read_gemini_env().expect("read live env");
assert!(
!live.contains_key("GOOGLE_API_KEY"),
"the leaked credential must be gone from ~/.gemini/.env: {live:?}"
);
assert_eq!(
live.get("HTTPS_PROXY").map(String::as_str),
Some("http://127.0.0.1:7890"),
"a hand-added live-only var must survive targeted removal: {live:?}"
);
}
#[tokio::test]
#[serial]
async fn scrub_gemini_live_cleanup_preserves_the_rest_of_the_env_file() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
// 这是一次用户没主动触发的启动期清理，不该顺手重写与泄漏无关的内容。
// read→HashMap→write 的往返会把注释、空行、无法识别的行全丢掉并按键名重排。
let original = "\
# my own notes
GOOGLE_API_KEY=key-C-owned
GOOGLE_API_KEY=key-A-leaked
this line is not KEY=VALUE at all
GEMINI_TIMEOUT_MS=30000
";
crate::gemini_config::write_gemini_env_text_atomic(original).expect("seed live env");
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("scrub must succeed");
let raw = std::fs::read_to_string(crate::gemini_config::get_gemini_env_path())
.expect("read live env");
assert!(
!raw.contains("key-A-leaked"),
"the leaked line must be gone: {raw:?}"
);
assert!(
raw.contains("# my own notes"),
"comments must survive a targeted removal: {raw:?}"
);
assert!(
raw.contains("this line is not KEY=VALUE at all"),
"unparseable lines must survive a targeted removal: {raw:?}"
);
// 被泄漏值遮住的那条重新生效——正是想要的结果，遮住它的恰恰是泄漏值
assert_eq!(
crate::gemini_config::read_gemini_env()
.expect("read live env")
.get("GOOGLE_API_KEY")
.map(String::as_str),
Some("key-C-owned"),
"only the matching line may be dropped: {raw:?}"
);
}
#[tokio::test]
#[serial]
async fn scrub_gemini_aborts_before_clearing_the_snippet_when_the_live_backup_fails() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
// 关代理时这份快照会被原样写回 live。若清不动它却照样清了片段、置了完成标记，
// 代理一停凭据就复活，而一次性标记保证不会再清第二次。
db.save_live_backup("gemini", "}not json{")
.await
.expect("seed backup");
let result = ProviderService::scrub_leaked_gemini_common_config(&state).await;
assert!(
result.is_err(),
"a backup that cannot be cleaned must abort the scrub"
);
// 片段是「该剥哪些键」的唯一知识来源，中止后必须原样留着，否则下次重试
// 会因为 poison 为空而直接短路，反倒把标记置上
let snippet = db
.get_config_snippet("gemini")
.expect("read snippet")
.expect("snippet must still exist");
assert!(
snippet.contains("key-A-leaked"),
"the snippet must be left intact so the next boot can retry: {snippet}"
);
assert!(
db.get_setting("gemini_common_config_credentials_scrubbed_v1")
.expect("read flag")
.is_none(),
"the one-shot flag must not be set when the scrub aborted"
);
}
#[tokio::test]
#[serial]
async fn scrub_gemini_leaves_no_residue_for_backfill_to_persist() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("scrub must succeed");
// 顺序陷阱回归：如果只清了片段，切走供应商时 remove_common_config_from_settings
// 就不再认识这个键，live 里的残留会被 backfill 永久写进供应商配置。
// 清理必须是原子的——清完之后，任何地方都不该再有那个值。
let snippet = db
.get_config_snippet("gemini")
.expect("read snippet")
.unwrap_or_default();
assert!(!snippet.contains("key-A-leaked"));
for (id, provider) in db.get_all_providers("gemini").expect("providers") {
assert!(
!provider
.settings_config
.to_string()
.contains("key-A-leaked"),
"provider '{id}' still carries the leaked value"
);
}
}
#[tokio::test]
#[serial]
async fn scrub_gemini_is_idempotent_and_skips_on_second_run() {
let _home = TempHome::new();
crate::settings::reload_settings().expect("reload settings");
let db = Arc::new(Database::memory().expect("init db"));
let state = AppState::new(db.clone());
seed_leaked_gemini_state(&db);
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("first run");
// 第二次必须是 no-op：用户清理后重新填的凭据不能被再抹一遍
db.set_config_snippet(
"gemini",
Some(json!({"GOOGLE_API_KEY": "restored"}).to_string()),
)
.expect("user re-adds a value");
ProviderService::scrub_leaked_gemini_common_config(&state)
.await
.expect("second run");
let snippet = db
.get_config_snippet("gemini")
.expect("read snippet")
.expect("snippet exists");
assert!(
snippet.contains("restored"),
"the one-shot flag must prevent a second scrub: {snippet}"
);
}
#[test]
fn extract_claude_common_config_strips_all_credentials_keeps_shareable() {
// env 混入多种凭据（Anthropic/OpenRouter/Google/OpenAI/Gemini + AWS/Vertex）
@@ -3029,20 +3492,39 @@ impl ProviderService {
/// `OPENROUTER_API_KEY` / `GOOGLE_API_KEY` 等回退）、各类 `*_AUTH_TOKEN` /
/// 单数 `*_TOKEN`、AWS Bedrock / Vertex 凭据、以及通用 secret / password /
/// 私钥命名。
fn is_sensitive_config_key(name: &str) -> bool {
pub(crate) fn is_sensitive_config_key(name: &str) -> bool {
let upper = name.to_ascii_uppercase();
// 单数 `_TOKEN` 命中 AWS_SESSION_TOKEN 等，但**不**误伤复数 `_TOKENS`
// （CLAUDE_CODE_MAX_OUTPUT_TOKENS / MAX_THINKING_TOKENS 是正常可共享配置）。
const SENSITIVE_SUFFIXES: &[&str] = &[
// 裸 `_KEY` 是最常见的凭据写法（OPENAI_KEY / GROQ_KEY / XAI_KEY…），
// 必须单列：只枚举 `_API_KEY` / `_ACCESS_KEY` 这些子类，等于把最普通
// 的那一种漏在外面。下面几条 `_*_KEY` 被它蕴含，保留是为了说明覆盖面。
"_KEY",
"_API_KEY",
"_APIKEY",
"_AUTH_TOKEN",
"_TOKEN",
"_ACCESS_KEY",
"_ACCESS_KEY_ID",
"_KEY_ID",
"_PRIVATE_KEY",
// 不带分隔符的复合写法各走各的后缀：`_KEY` 够不着 `..._APIKEY`
// （倒数第四个字符是 I 不是下划线）。VOLC_ACCESSKEY 是火山引擎文档
// 里的正式变量名，本仓库就实现了火山 AK/SK 用量查询。
"_APIKEY",
"_ACCESSKEY",
"_SECRETKEY",
"_APITOKEN",
"_AUTH_TOKEN",
"_TOKEN",
// GITHUB_PAT / GITLAB_PAT 等 personal access token 的惯用写法，
// 既不含 TOKEN 也不含 KEY，前面每一条规则都够不着。
"_PAT",
// 口令类的常见缩写。`_PASS` 不会误伤 `*_BYPASS`（那个以 `_BYPASS`
// 结尾），`_PWD` 也不会误伤 shell 的 PWD / OLDPWD。
"_PWD",
"_PASS",
"_PASSPHRASE",
"_CREDS",
];
const SENSITIVE_EXACT: &[&str] = &[
"APIKEY",
@@ -3242,7 +3724,13 @@ impl ProviderService {
let mut snippet = serde_json::Map::new();
if let Some(env) = env {
for (key, value) in env {
if key == "GOOGLE_GEMINI_BASE_URL" || key == "GEMINI_API_KEY" {
// 端点按名剥离（它不是凭据，模式匹配够不着）；凭据全部交给
// `is_sensitive_config_key` 统一模式匹配（与 Claude 提取器一致）。
// 只列固定名单会漏掉下一个 `*_API_KEY` —— 例如 `GOOGLE_API_KEY`
// （provider.rs 认可的一等 Gemini 凭据），而共享片段会被 deep-merge
// 回其它 Gemini 供应商，漏剥即等于把 A 账号的密钥写进 B 供应商并
// 发往 B 的 base_url。`GEMINI_API_KEY` 不必单列：`_KEY` 后缀已覆盖。
if key == "GOOGLE_GEMINI_BASE_URL" || Self::is_sensitive_config_key(key) {
continue;
}
let Value::String(v) = value else {
@@ -3263,6 +3751,221 @@ impl ProviderService {
.map_err(|e| AppError::Message(format!("Serialization failed: {e}")))
}
/// 一次性清理：把历史泄漏进 Gemini 共享片段的凭据从所有存储位置抹掉。
///
/// 背景：`extract_gemini_common_config` 曾只剥离两个固定键名，`GOOGLE_API_KEY`
/// 等一等凭据会进入共享片段，再被 `apply_common_config_to_settings` 深合并进
/// **其它** Gemini 供应商的 env，随请求发往对方的 base_url。
///
/// 光修提取器不够：Gemini 的片段一旦生成就**永不自动重提取**（启动期
/// auto-extract 与导入后补提取都要求 `snippet.is_none()`，切换时的回写又只对
/// Claude / Codex 生效），所以存量片段会一直带着密钥继续注入。
///
/// 两个关键约束：
///
/// 1. **不能只清片段**。合并与剥离是一对靠「值相等」严格抵消的操作：切走供应商时
/// `remove_common_config_from_settings` 依据片段内容把注入的键删掉。片段里一旦
/// 没了这个键，backfill 就会把 live 中残留的密钥原样写进受害供应商的
/// `settings_config`——泄漏从瞬时污染变成永久污染。所以片段、各供应商配置、
/// live 文件必须一起清。
/// 2. **按值相等定向删除，不按键名一刀切**。复用 `remove_common_config_from_settings`
/// 可以只清掉扩散出去的那一份，保留某个供应商自己写的、值不同的同名键。
///
/// 步骤顺序本身是安全属性的一部分：**清片段必须排在最后**。片段是
/// `remove_common_config_from_settings` 唯一的"该剥哪些键"来源，一旦清空，任何
/// 残留（live 文件里的、下一轮重试要处理的）都再也无法被识别和剥离。所以所有
/// 可能失败的步骤都排在它前面，失败即带错返回，让下次启动能原样重来。
///
/// 清理后部分供应商会显示缺少 API Key，需用户重填——这是正确行为：那把密钥本就
/// 不属于它们。（受害者原有的同名键在合并时已被覆盖，无法恢复。）动手前会往
/// settings 的 `gemini_common_config_scrub_audit_v1` 写一条审计记录，内容是
/// **键名与受影响的供应商 id，不含值**：`settings` 会随 WebDAV/S3 同步上传，
/// 而这里处理的正是必须销毁的凭据，留值等于把一次清除换成一份跨设备扩散、
/// 没有界面入口、永不过期的明文副本。
pub async fn scrub_leaked_gemini_common_config(state: &AppState) -> Result<(), AppError> {
const FLAG: &str = "gemini_common_config_credentials_scrubbed_v1";
const AUDIT_KEY: &str = "gemini_common_config_scrub_audit_v1";
let app = AppType::Gemini;
if state.db.get_bool_flag(FLAG).unwrap_or(false) {
return Ok(());
}
let Some(snippet_text) = state.db.get_config_snippet(app.as_str())? else {
state.db.set_setting(FLAG, "true")?;
return Ok(());
};
// 片段解析不了就不动它，只标记完成——乱改用户数据比留着更糟
let Ok(Value::Object(entries)) = serde_json::from_str::<Value>(&snippet_text) else {
state.db.set_setting(FLAG, "true")?;
return Ok(());
};
let mut poison = serde_json::Map::new();
let mut clean = serde_json::Map::new();
for (key, value) in entries {
if Self::is_sensitive_config_key(&key) {
poison.insert(key, value);
} else {
clean.insert(key, value);
}
}
if poison.is_empty() {
state.db.set_setting(FLAG, "true")?;
return Ok(());
}
log::warn!(
"检测到 {} 个凭据键残留在 Gemini 通用配置片段中，开始一次性清理",
poison.len()
);
let poison_keys: Vec<String> = poison.keys().cloned().collect();
let poison_value = Value::Object(poison);
let poison_text = serde_json::to_string(&poison_value)
.map_err(|e| AppError::Message(format!("Serialization failed: {e}")))?;
// 1) 先算出各供应商清理后的配置，但**先不落库**
let providers = state.db.get_all_providers(app.as_str())?;
let mut pending: Vec<(String, Provider, Value)> = Vec::new();
for (id, provider) in providers {
let cleaned = match live::remove_common_config_from_settings(
&app,
&provider.settings_config,
&poison_text,
) {
Ok(cleaned) => cleaned,
Err(err) => {
log::warn!("清理供应商 '{id}' 的泄漏凭据失败: {err}");
continue;
}
};
if cleaned != provider.settings_config {
pending.push((id, provider, cleaned));
}
}
// 2) 落库前留一份审计记录：**只记键名与受影响的供应商，不记值**。
//
// 「按值相等定向删除」在一种合法场景下也会命中：用户有意在多个供应商里
// 复用同一把 key。所以必须留下"删了什么、从哪删的"，否则用户只能靠翻
// 日志。但不能留值——`settings` 表不在 `SYNC_SKIP_TABLES` 里，会随
// WebDAV/S3 同步上传，而这里处理的恰恰是必须销毁的泄漏凭据：留值等于
// 把一次清除换成一份没有界面入口、永不过期、还会跨设备扩散的明文副本。
// 密钥本来就该轮换，可恢复性不值这个代价。
let removed_env_keys = |before: &Value, after: &Value| -> Vec<String> {
let before_env = before.get("env").and_then(Value::as_object);
let after_env = after.get("env").and_then(Value::as_object);
match (before_env, after_env) {
(Some(before_env), Some(after_env)) => before_env
.keys()
.filter(|key| !after_env.contains_key(*key))
.cloned()
.collect(),
(Some(before_env), None) => before_env.keys().cloned().collect(),
_ => Vec::new(),
}
};
let audit = serde_json::json!({
"removedFromSnippet": poison_keys,
"providers": pending
.iter()
.map(|(id, provider, cleaned)| serde_json::json!({
"id": id,
"removedKeys": removed_env_keys(&provider.settings_config, cleaned),
}))
.collect::<Vec<_>>(),
});
let audit_text = serde_json::to_string(&audit)
.map_err(|e| AppError::Message(format!("Serialization failed: {e}")))?;
// 只在没有记录时写。provider 的写入不是一个事务（每次 save_provider 各自
// 提交），上一轮可能改到一半就中止；此时完成标记没置位，下次启动会重跑，
// 而重跑看到的"原始状态"已经残缺。无条件 INSERT OR REPLACE 会拿这份残缺
// 记录盖掉第一轮那份完整的。
if state.db.get_setting(AUDIT_KEY)?.is_none() {
state.db.set_setting(AUDIT_KEY, &audit_text)?;
}
// 3) 各供应商 settings_config：按值相等定向删除扩散出去的副本
for (id, provider, cleaned) in pending {
let mut updated = provider;
updated.settings_config = cleaned;
state.db.save_provider(app.as_str(), &updated)?;
log::info!("已从 Gemini 供应商 '{id}' 中清除泄漏的共享凭据");
}
// 4) 代理接管中的 live 快照里也可能有一份副本。这一步的失败**必须传播**：
//
// 关代理时 `restore_live_config_for_app_with_fallback_inner`（proxy.rs:869）
// 会把这份快照原样写回 `~/.gemini/.env`。若它仍带毒而我们照样清了片段、置了
// 完成标记，那么代理一停凭据就当场复活，而一次性标记又保证不会再清第二次；
// 此后片段里已没有这个键，下一次切换的 backfill 就把它永久写进受害供应商的
// 配置——还是本函数开头那个顺序陷阱，只是换了扇门进来。
//
// 带错返回是安全的失败方式：调用方（lib.rs:1189）只记 warn 不中断启动，
// 片段和标记都原样留着，下次启动照原样重来。
if let Some(backup) = state.db.get_live_backup(app.as_str()).await? {
let original: Value = serde_json::from_str(&backup.original_config)
.map_err(|e| AppError::Message(format!("解析 Gemini 代理接管备份失败: {e}")))?;
let cleaned = live::remove_common_config_from_settings(&app, &original, &poison_text)?;
if cleaned != original {
let text = serde_json::to_string(&cleaned)
.map_err(|e| AppError::Message(format!("Serialization failed: {e}")))?;
state.db.save_live_backup(app.as_str(), &text).await?;
log::info!("已从 Gemini 代理接管备份中清除泄漏的共享凭据");
}
}
// 5) `~/.gemini/.env`：**定向**删除，且必须在清片段之前做，失败即中止。
//
// 为什么不用 `sync_current_provider_for_app` 重投影：它在没有当前供应商
// 时直接返回 Ok 而根本不写文件，泄漏值会原样留在 live 里；等片段被清空
// 之后，下次切换时 `remove_common_config_from_settings` 再也认不出这个
// 键，backfill 就把它永久写进受害供应商的配置——正是本函数开头说的那个
// 顺序陷阱，只是由"没修"变成"修了一半更糟"。定向删除还顺带保住了只存在
// 于 live、与供应商无关的手工 env（重投影会把它们抹掉）。
//
// 删除走 `remove_gemini_env_entries` 的**保序**实现而不是 read→HashMap→
// write 往返：后者会顺手抹掉注释、空行和无法识别的行，并按键名重排整个
// 文件。全量投影时那无所谓，但这里是一次用户没主动触发的启动期清理，不该
// 连带改写与泄漏无关的内容。
//
// 失败就带着错误返回：片段此刻还留着毒键，完成标记也没置位，下次启动能
// 照原样重来。清片段是不可逆的一步，必须排在所有会失败的步骤之后。
let poison_env: HashMap<String, String> = poison_value
.as_object()
.map(|map| {
map.iter()
.filter_map(|(key, value)| {
value.as_str().map(|text| (key.clone(), text.to_string()))
})
.collect()
})
.unwrap_or_default();
if crate::gemini_config::remove_gemini_env_entries(&poison_env)? {
log::info!("已从 ~/.gemini/.env 中清除泄漏的共享凭据");
}
// 6) 片段本身：保留可共享的部分。全部清空时删行而不是写 "{}"——留着空行会让
// should_auto_extract_config_snippet 永远为 false，用户的合法共享配置再也
// 重建不回来。同理绝不置 cleared 标记。
if clean.is_empty() {
state.db.set_config_snippet(app.as_str(), None)?;
} else {
let cleaned_snippet = serde_json::to_string_pretty(&Value::Object(clean))
.map_err(|e| AppError::Message(format!("Serialization failed: {e}")))?;
state
.db
.set_config_snippet(app.as_str(), Some(cleaned_snippet))?;
}
state.db.set_setting(FLAG, "true")?;
log::info!("Gemini 通用配置凭据清理完成");
Ok(())
}
/// Extract common config for OpenCode (JSON format)
fn extract_opencode_common_config(settings: &Value) -> Result<String, AppError> {
// OpenCode uses a different config structure with npm, options, models
src-tauri/src/services/skill.rs
+1451 -139
View File
File diff suppressed because it is too large Load Diff