- `bl mcp list` — list MCP servers enabled under the current Bailian
account via console gateway PageList (always activated=1).
- `bl mcp tools <server-code>` — list tools exposed by a server.
- `bl mcp call <server-code>.<tool>` — invoke a tool. Accepts `--json`,
repeatable `--arg k=v` (JSON-parsed when possible) and `--query` sugar;
`--url` overrides the endpoint for non-Bailian MCPs.
- core: export `bailianMcpUrl(baseUrl, code)` building
`/api/v1/mcps/<code>/mcp`; `McpClient` now takes a full URL.
- `bl search web` switches to `mcpWebSearchEndpoint` directly.
- e2e: `mcp.e2e.test.ts` covering help, dry-run, arg-merge semantics,
invalid-input paths, and one live `tools/list` against WebSearch.
script/js executes its `code` as host JavaScript (via new Function), and a step's
`code` is a *resolved* input — so it could be written as `{ $from: <chat-step> }`,
turning model/API output into the body of the executed function (untrusted data
-> arbitrary host code execution). Pipeline validation now requires script/js
`code` to be a literal string: any $from/expression-sourced code is rejected.
Authoring a literal script/js step remains supported (the pipeline file is the
trust boundary, like a shell/npm script). Combined with "dry-run never executes
$js", this closes the path where untrusted text reaches the JS sink.
Adds regression tests: $from-sourced code rejected, literal code accepted,
dry-run does not execute $js, getByJsonPointer blocks prototype/inherited keys,
and concurrency clamps to the maximum.
https://claude.ai/code/session_017ZGQCjwNQF5Pz96gLUnnG1
若麒