chore(infra): build/deps/MCP/whitelist fixes (split 1/3 from #1057)#1059
Open
tech-synity wants to merge 7 commits into
Open
chore(infra): build/deps/MCP/whitelist fixes (split 1/3 from #1057)#1059tech-synity wants to merge 7 commits into
tech-synity wants to merge 7 commits into
Conversation
The bare pattern `pkg-helper` on line 6 matched any file or directory named pkg-helper anywhere in the tree — including the tracked source directory `cmd/pkg-helper/`. Git tolerated this because already-tracked files override ignore rules, but uploaders that apply .gitignore patterns naively to the filesystem (notably `railway up`) stripped the source directory from the build context, breaking `go build ./cmd/pkg-helper` in Docker with `stat /src/cmd/pkg-helper: directory not found`. Anchor both binary-artifact patterns (`/openclaw-go`, `/pkg-helper`) to the repo root so they only match the compiled output, not the source. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Working tree kept surfacing `goclaw.exe` / `goclaw-local.exe` (Windows dev builds) and ad-hoc `tmp-reset-bot/` probe scripts as untracked. A single `git add .` would commit them — both are useless outside the author's machine and the debug probes import `internal/` packages from outside `cmd/` (ugly but legal), so they should never ride into Git. Add to .gitignore (anchored like the existing `/openclaw-go` entry, so matching doesn't accidentally touch source dirs — same concern that broke Docker builds in 2de99e26): /goclaw.exe /goclaw-local.exe /tmp-reset-bot/ /tmp-probe*/ Also add .railwayignore file — new; Railway copies the working tree verbatim and already ignored tests/node_modules/etc. Add `tmp-*` glob so ad-hoc probe dirs don't bloat image layers on `railway up`. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Routine dependency refresh. v2.12 pulls in a new indirect dep (git.sr.ht/~jackmordaunt/go-toast/v2 v2.0.3 — used by Wails' notification path) and otherwise contains only maintenance fixes from upstream. Verified: `go build -tags sqliteonly ./internal/... ./cmd/` green — desktop edition's Wails bindings still compile. No goclaw code calls the notification API directly, so the new indirect is latent unless Wails decides to pull it into the main path later. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
… in strict mode
OpenAI strict mode rejected `use_skill` calls with:
invalid_function_parameters: 'additionalProperties' is required to be
supplied and to be false
at path ('properties', 'params', 'type', '0')
Root cause: applyStrictMode early-returned when a node was
`{"type":"object"}` with no nested `properties` key, leaving
additionalProperties unset. Then makeNullable widened the type to
`["object","null"]`, and the strict validator checks each branch
independently — the "object" variant was missing additionalProperties
so the whole schema failed validation.
Fix: when typ=="object", set additionalProperties:false even if the
node has no inner properties. This covers the common "bag of
free-form params" shape tool authors write as:
params: { type: "object", description: "..." }
Trade-off worth documenting: strict mode + bare object ⇒ OpenAI will
reject any call that actually populates those params. If a tool truly
needs a free-form dict of arbitrary keys, strict mode is the wrong
fit — either declare the known keys under `properties`, or opt that
tool out of strict. Left a comment on the branch spelling this out.
Regression test added: TestApplyStrictMode_BareObjectProperty asserts
both `additionalProperties:false` AND the `["object","null"]` type
union survive the transform.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Previously, builds without the `embedui` tag left "/" unhandled, so http.ServeMux returned a bare 404 when an operator opened the deployed URL in a browser. That's needlessly opaque — it looks like the service is broken even though the gateway is healthy. Register a small JSON index handler for "/" in the no-UI branch. It returns service status + protocol version + a list of real endpoints for an exact "/" match, and a JSON 404 for any other unmatched path (http.ServeMux routes "/" as a catch-all). Embedui builds are unchanged — the webui.Handler() still owns "/" and takes precedence. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Lets admins attach server-specific quirks (e.g. "no trailing semicolons in code args") to MCP tool descriptions without modifying the upstream MCP server or redeploying goclaw. Hints are stored in the existing mcp_servers.settings JSONB (no migration) following the same pattern as require_user_credentials. Backend: - Add mcp.ToolHints type + ParseToolHints() helper alongside requireUserCreds - Add BridgeTool.WithHints(global, toolHint) setter that appends a suffix to Description() so the LLM sees hints as part of the tool schema - Thread hints through connectServer / connectViaPool / loop_mcp_user - 12 new unit tests for ParseToolHints + WithHints (+ edge cases) Frontend: - New "AI Agent Hints (Optional)" section in MCP form dialog: a global textarea + a key-value list for per-tool hints - Extend KeyValueEditor with valueAs="textarea" prop (reusable) - Zod schema + TS types + i18n (en/vi/zh) updated Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…cake parity) Creating a Bitrix24 channel_instance failed with "invalid channel_type" even though Phase 03 wired the factory and the UI added Bitrix24 to the picker. The write path hits two parallel whitelist switches — one for WS RPC in internal/gateway/methods and one for HTTP in internal/http — and both missed bitrix24. The WS switch also silently missed facebook and pancake, so anything driven through the WS API was rejecting channels the HTTP API already accepted. - Add bitrix24 to both switches. - Add facebook and pancake to the WS switch so the two functions stop drifting. CHANNEL_TYPES in the UI constants file was already offering both. - Annotate each switch with a keep-in-sync comment pointing at its twin and at ui/web/src/constants/channels.ts. The previous drift was silent (untyped string list across three separate files); the comment at least makes the next channel addition noisy enough to catch. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
want full mcp support. now mcp resource miss |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
PR 1/3 split from #1057 per maintainer review. Channel-agnostic infrastructure fixes only — safe to merge first; PRs 2 (store + migration) and 3 (Bitrix24 channel) follow.
Commits (7)
Why split
Maintainer review on #1057 flagged MEGA-PR risk. These commits touch unrelated infrastructure (deps, build, MCP feature, OpenAI strict mode) — independently reviewable, low risk to revert. The channel_type whitelist update is preparatory but does not require Bitrix24 code to function (just allows the type string).
Test plan
Coordination