mirror of
https://github.com/openclaw/openclaw.git
synced 2026-10-03 09:39:25 +00:00
docs/gateway/config-tools.md was 59,564 characters across two H2 topics that mixed reference keys, a long explanation of the managed GitHub identity, and how-to provider recipes. Split by reader job into docs/gateway/config-tools/, keeping the parent as a short index. Children: - tool-policy — tool profiles, tool groups, the sandbox tool gate, tools.codeMode, and the allow/deny, byProvider, toolsBySender, and elevated layers (14,547 chars) - github-identity — tools.github: the shared managed gh identity, token refresh, and the execution boundaries that receive it (10,641 chars) - built-in-tools — tools.exec, tools.loopDetection, tools.web, tools.media, tools.updatePlan (6,254 chars) - sessions-and-subagents — tools.agentToAgent, tools.sessions, tools.sessions_spawn, agents.defaults.subagents (7,033 chars) - custom-providers — models.providers registration, base-URL trust, and the provider field reference (12,309 chars) - provider-examples — worked configurations for nine providers (8,980 chars) Anchor strategy: per-anchor routes are impossible here because redirectSource() rejects any source containing [?#], so every id the page published stays alive on the parent index as an authored <a id="..." /> stub next to a link to its new home. parseDocsDocument reports 58 ids before the split; 57 are stubbed and `related` is still published by the index itself, so no duplicate authored/canonical ID is raised. Backticked headings emit both a dotted and a dashed id (tools.github and tools-github) and both are preserved, as is the percent-encoded tools.allow-%2F-tools.deny, the pre-existing authored toolsupdateplan stub, and the ten Accordion-minted ids. All 58 resolve through resolveDocsFragment against the new index, and all 40 index-to-child fragment links resolve on their destination child. Losslessness (source body vs. the six children, measured mechanically against the base sha, not against a working copy): 54 fence markers before and after with identical info strings; 47 table rows before and after; all 12 original outbound links still present; each child's body byte-identical to its source line range after H3-to-H2 promotion. Two thematic breaks that separated the former H2 sections are dropped because those sections are now separate pages, and the empty container heading "## Tools" is not reproduced — its anchor is stubbed. One prose change, the declared orphaned-cross-reference exception: "See the llama.cpp example below" in the compat table pointed at a section that now lives on another page, so it became a real link. The other six directional references stayed within a single child and were re-verified there. taxonomy.yaml references this page without a fragment, so validateTaxonomyDocsReferences still resolves it; maturity:check passes. The eight qa/scenarios docsRefs and the two src/ docs URLs also target the surviving route. Closes audit findings: r3-0337, r3-1539
7.3 KiB
7.3 KiB
| summary | read_when | title | sidebarTitle | |||
|---|---|---|---|---|---|---|
| Tools config (policy, experimental toggles, provider-backed tools) and custom provider/base-URL setup |
|
Configuration — tools and custom providers | Tools and custom providers |
tools.* config keys and custom provider / base-URL setup. For agents, channels, and other top-level config keys, see Configuration reference.
What each page covers
- Configuration — tool policy — tool profiles, tool groups, the sandbox tool gate,
tools.codeMode, and theallow/deny,byProvider,toolsBySender, andelevatedlayers. - Configuration — GitHub identity for agent tools —
tools.github: the shared managedghidentity, token refresh, and the execution boundaries that receive it. - Configuration — built-in tool settings —
tools.exec,tools.loopDetection,tools.web,tools.media, andtools.updatePlan. - Configuration — cross-agent, session, and subagent tools —
tools.agentToAgent,tools.sessionsvisibility,tools.sessions_spawnattachments, andagents.defaults.subagents. - Configuration — custom providers and base URLs —
models.providersregistration, base-URL trust, and the provider field reference. - Configuration — provider examples — worked configurations for Cerebras, Kimi, llama.cpp, LM Studio, MiniMax, Moonshot, OpenCode, Synthetic, and Z.AI.
Where each section moved
Every heading this page used to publish keeps its anchor here, so an existing
link such as /gateway/config-tools#tools-agenttoagent still resolves. Each
entry points at the page that now holds the content.
- Tools
- Tool profiles
- Tool groups
- MCP and plugin tools inside sandbox tool policy
tools.codeModetools.allow/tools.denytools.byProvidertools.toolsBySendertools.elevatedtools.githubtools.exectools.loopDetectiontools.webtools.media- Media model entry fields
tools.updatePlantools.agentToAgenttools.sessions- Visibility scopes
tools.sessions_spawn- Attachment notes
agents.defaults.subagents- Custom providers and base URLs
- Auth and merge precedence
- Provider field details
- Top-level catalog
- Provider connection and auth
- Request transport overrides
- Model catalog entries
- Custom provider capability declarations
- Amazon Bedrock discovery
- Provider examples
- Cerebras (GLM 4.7 / GPT OSS)
- Kimi Coding
- Local models (llama.cpp / llama-server)
- Local models (LM Studio)
- MiniMax M3 (direct)
- Moonshot AI (Kimi)
- OpenCode
- Synthetic (Anthropic-compatible)
- Z.AI (GLM-4.7)
Related
- Configuration — agents
- Configuration — channels
- Configuration reference — other top-level keys
- Tools and plugins