LibreChat/api/server/services/Endpoints/agents/initialize.js
Danny Avila 24e29aa8cb
🌱 fix: Inject Code-Tool Files Into Graph Sessions on First Call (+ read_file Sandbox Fallback) (#12831)
* 🌱 fix: Seed Code Tool Files Into Graph Sessions on First Call

Files attached to an agent's `tool_resources.execute_code` (user uploads
or generated artifacts from a prior turn) were silently dropped on the
first `execute_code` invocation of a turn. The agents-side `ToolNode`
populates `_injected_files` only when its `sessions` map already has an
`EXECUTE_CODE` entry — but that entry is only written by a previous
successful execution, so call #1 had nothing to inject. CodeExecutor
then fell back to a `/files/{session_id}` fetch, but `session_id` was
also empty on call #1, leaving the sandbox without the primed files.

Mirror the existing skill-priming pattern (`primeInvokedSkills` →
`initialSessions`) for code-resource files: eagerly call `primeFiles`
before `createRun` and merge the result into `initialSessions` via a
new `seedCodeFilesIntoSessions` helper. Skill files and code-resource
files now share the same `EXECUTE_CODE` entry; the prior representative
`session_id` is preserved on merge.

* 🔬 chore: Add Diagnostic Logging for Code-Files Seeding

Temporary debug logs to diagnose why first-call file injection is not
firing in real agent runs. Logs `wantsCodeExec`, available tool-resource
keys, primed file count, and the seeded EXECUTE_CODE entry. Will revert
once the failure mode is identified.

* 🪛 refactor: Capture primedCodeFiles per-agent at init, merge across run

Replace the client.js eager `primeFiles` call with a per-agent capture at
initialization time so every agent in a multi-agent run (primary +
handoff + addedConvo) contributes its `tool_resources.execute_code`
files to the shared `Graph.sessions` seed.

- handleTools.js (eager loadTools): the `execute_code` factory closes
  over a `primedCodeFiles` slot and surfaces it in the return.
- ToolService.js loadToolDefinitionsWrapper (event-driven): captures
  `files` from the existing `primeCodeFiles` call (was dropping them
  while only keeping `toolContext`) and surfaces them.
- packages/api initialize.ts: the loadTools callback contract now
  includes `primedCodeFiles`, threaded onto `InitializedAgent`.
- client.js: iterate `[primary, ...agentConfigs.values()]` and merge
  each agent's `primedCodeFiles` into `initialSessions`. Drop the
  primary-only `primeCodeFiles` call and diagnostic logs from the prior
  attempt — wrong layer (single-agent), wrong gate (`agent.tools`
  contained Tool instances after init, so the `.includes("execute_code")`
  string check always failed).

* 🔬 chore: Add per-agent diagnostic logs for code-files seeding

Logs `tool_resources` keys + file counts inside loadToolDefinitionsWrapper
and per-agent `primedCodeFiles` + final initialSessions inside
AgentClient. Will revert once the failure mode is confirmed.

* 🔬 chore: Add file-lookup diagnostics inside initializeAgent

Logs the inputs and intermediate counts of the conversation-file lookup
chain (convo file ids, thread message ids, code-generated and
user-code file counts) so we can pinpoint why `tool_resources.execute_code`
is arriving empty at `loadToolDefinitionsWrapper` despite the agent
having `execute_code` in its tools list.

* 🔬 chore: Probe execute_code files without messageId filter

Adds a relaxed `getFiles({conversationId, context: execute_code})` probe
that runs only when `getCodeGeneratedFiles` returns empty. Lists what's
actually in the DB for this conversation so we can confirm whether the
file is missing entirely or whether the messageId filter is rejecting it.

* 🔬 chore: Fix probe getFiles arg order (sort vs projection)

Probe was passing a projection object as the sort arg, which mongoose
rejected with `Invalid sort value`. Move it to the third arg
(selectFields) so the probe actually runs.

* 🪢 fix: Preserve Original messageId on Code-Output File Update

Each `processCodeOutput` call was overwriting the persisted file's
`messageId` with the *current* run's id. When a turn re-creates an
existing file (filename + conversationId match → `claimCodeFile`
returns the existing record, `isUpdate=true`), the file's link to
the assistant message that originally produced it gets clobbered.

`initializeAgent` later runs `getCodeGeneratedFiles({messageId: $in: <thread>})`
to seed `tool_resources.execute_code` from prior-turn artifacts. With a
stale `messageId` (e.g. from a failed read attempt that re-shelled the
same filename), the file no longer matches the parent-walk thread, so
`tool_resources` arrives empty at agent init, the new
`primedCodeFiles` channel has nothing to seed, and the LLM can't see
its own prior-turn artifacts on the next turn — defeating the
just-added Graph-sessions seeding fix.

Preserve the existing `claimed.messageId` on update; first-creation
behavior is unchanged. The runtime return value still includes the
current run's `messageId` (via `Object.assign(file, { messageId })`)
so the artifact is correctly attributed to the live tool_call.

* 🧹 chore: Remove diagnostic logs from code-files seeding path

Drops the temporary debug logs added to trace the empty-tool_resources
failure mode. Production code paths (loadToolDefinitionsWrapper,
client.js seed loop, initializeAgent file lookup) are left as the
permanent shape: capture primedCodeFiles, merge across agents, seed
initialSessions before run start.

* 🪛 feat: read_file Sandbox Fallback for /mnt/data + Non-Skill Paths

When the model called `read_file` with a code-execution path (e.g.
`/mnt/data/sentinel.txt`), the handler returned a misleading
`Use format: {skillName}/{path}` error. Adds a sandbox-aware fallback:

- Short-circuit `/mnt/data/...` (can never be a skill reference) →
  route to a sandbox `cat` via the new host-provided `readSandboxFile`
  callback, which POSTs to the codeapi `/exec` endpoint.
- Skip the skill resolver entirely when `accessibleSkillIds` is empty
  — the resolved-output of `resolveAgentScopedSkillIds` already
  collapses the admin capability + ephemeral badge + persisted
  `skills_enabled` chain, so an empty value is the authoritative
  "skills aren't in scope for this agent" signal.
- For `{firstSegment}/...` paths, consult the catalog-derived
  `activeSkillNames` Set (no DB read) to detect non-skill names and
  fall through to the sandbox before the model has to retry with
  `bash_tool`.

`activeSkillNames` is captured from `injectSkillCatalog`, threaded onto
`InitializedAgent`, into `agentToolContexts`, then through
`enrichWithSkillConfigurable` into `mergedConfigurable` for the handler.

The host implementation of `readSandboxFile` lives in
`api/server/services/Files/Code/process.js` and shells `cat <path>`
through the seeded sandbox session — `tc.codeSessionContext`
(emitted by ToolNode for `read_file` calls in `@librechat/agents`
v3.1.72+) provides the `session_id` + `_injected_files` so the read
lands in the same sandbox that holds prior-turn artifacts. When the
seeded context isn't available (older agents version, no codeapi
configured), the handler returns a model-visible error pointing at
`bash_tool` instead of silently failing.

Tests: 8 new `handleReadFileCall` cases cover the new short-circuits,
the skills-not-enabled gate, the activeSkillNames lookup, the
sandbox-fallback success path, and the bash_tool retry hint on
fallback failure. Existing `read_file` tests now opt into "skills are
in scope" via a `skillsInScope()` fixture (production wouldn't reach
the skill lookup with empty `accessibleSkillIds`).

* 🔧 chore: Update @librechat/agents dependency to version 3.1.72

Bumps the version of the @librechat/agents package across package-lock.json and relevant package.json files to ensure compatibility with the latest features and fixes.

* 🪛 refactor: Centralize Tool-Session Seed in buildInitialToolSessions Helper

Addresses review feedback on the per-agent merge in client.js:

- **Run-wide semantics, named explicitly.** The merge into a single
  `Graph.sessions[EXECUTE_CODE]` was a deliberate match to the
  agents-library design (`Graph.sessions` is shared across every
  `ToolNode` in the run), but the inline `for (const a of agents)`
  loop in `AgentClient.chatCompletion` made it look per-agent. Move
  the logic to a TS helper `buildInitialToolSessions` that documents
  the run-wide-by-design contract in one place. The CJS controller
  now contains a single call site, no business logic.

- **Subagent walk (P2).** The previous loop only iterated
  `[primary, ...agentConfigs.values()]`. Pure subagents are pruned
  out of `agentConfigs` after init and retained on each parent's
  `subagentAgentConfigs`, so their primed code files were silently
  dropped from the seed. The helper now walks recursively, with a
  visited-Set keyed on object identity that terminates safely on a
  malformed agent graph (cycle).

- **`jest.setup.cjs` polyfill for undici `File`.** Reviewer hit
  `ReferenceError: File is not defined` running the targeted spec on
  WSL — a known Node 18 issue where `globalThis.File` from
  `node:buffer` isn't auto-exposed. Polyfill it inside a Jest setup
  file so the suite boots regardless of Node patch version.

Helper test coverage (8 new): skill-only / agent-only / both,
recursive subagent walk, cycle-safe walk, primary+subagent
deduplication, undefined/null entries in the agents iterable, and
representative session_id preservation across the merge.

16 tests pass total in `codeFilesSession.spec.ts` (8 prior + 8 new).
No behavior change vs. the previous commit for the existing
primary+agentConfigs case — subagent inclusion is the only new
behavior, and it matches what the existing seeding logic would have
done if subagents had been in `agentConfigs`.

* 🪛 fix: FIFO Walk Order in buildInitialToolSessions (P3 review)

The traversal used `Array.pop()` (LIFO), which visited the LAST
top-level agent first. The docstring says "primary first"; the code
contradicted it. When no skill seed exists the first-visited agent's
first file supplies the representative `session_id` written to
`Graph.sessions[EXECUTE_CODE]` — so a LIFO walk silently flipped which
agent that came from. `ToolNode` ultimately uses per-file `session_id`s
for runtime injection (so behavior was indistinguishable for current
callers), but the discrepancy was a footgun for any future consumer
that read the representative.

Switch to FIFO via `Array.shift()` to match both the docstring and the
existing `loadSubagentsFor` walk pattern in
`Endpoints/agents/initialize.js`. Add a regression test that asserts
the primary's `session_id` is the representative (and that all three
agents' files still contribute, with per-file `session_id`s preserved).

* 🔬 test: Lock In Code-Files Bug Fixes Per Comprehensive Review

Addresses MAJOR + MINOR + NIT findings from the multi-pass review:

**Finding #4 (MINOR) — empty relativePath misses sandbox fallback.**
A model calling `read_file("output/")` where "output" isn't a skill
name dead-ended with `Missing file path after skill name` instead of
being routed to the sandbox like every other malformed-path branch.
Add the same `codeEnvAvailable → handleSandboxFileFallback` pattern,
plus two regression tests.

**Finding #7 (NIT) — duplicate `skillsInScope()` helper.**
Hoist the identical helper out of two nested describe blocks to
module scope. Single source of truth.

**Finding #1 (MAJOR) — `persistedMessageId` had zero test coverage.**
The fix preserves a file's original `messageId` on update so
`getCodeGeneratedFiles` can still match it on subsequent turns. A
regression in the `isUpdate ? (claimed.messageId ?? messageId) : messageId`
ternary would silently re-introduce the original cross-turn priming
bug. Five new tests cover:
- UPDATE preserves `claimed.messageId` in the persisted record
- UPDATE falls back to current run id when `claimed.messageId` is
  absent (legacy records predating the field)
- CREATE uses current run id (no claimed record exists)
- The runtime return value uses the LIVE id (artifact attribution)
  even when the persisted record kept the original
- The image branch follows the same contract (would silently regress
  if the ternary diverged across the two file-build branches)

The tests use a `snapshotCreateFileArgs()` helper because
`processCodeOutput` mutates the file object after `createFile`
returns (`Object.assign(file, { messageId, toolCallId })`) and a
naive `createFile.mock.calls[0][0]` would reflect the post-mutation
state instead of what was actually persisted.

**Finding #2 (MAJOR) — `readSandboxFile` had no direct tests.**
The model-controlled `file_path` flows through a POSIX single-quote
escape into a shell `cat` command, making this a security boundary.
A quoting regression would let a malicious filename break out of the
quoted argument and inject arbitrary shell. 20 new tests across:
- Shell quoting (7): plain filenames, embedded `'`, `$()`, backticks,
  newlines, shell metachars, multiple consecutive single-quotes
- Payload shape (6): /exec URL, bash language, conditional
  session_id / files inclusion, dedicated keepAlive:false agents
- Response handling (6): `{content}` on success, null on missing
  base URL or absent stdout, throws on stderr-only, partial-success
  returns stdout, transport errors are logged then rethrown
- Timeout (1): matches processCodeOutput's 15s SLA

Audited findings #5 (acknowledged tech debt — readSandboxFile in JS
workspace), #6 (pre-existing positional-args debt on
enrichWithSkillConfigurable), and #8 (cosmetic JSDoc style) — no
action taken per the reviewer's own assessment.

Audited finding #3 (walk order vs docstring) — already addressed in
commit 007f32341 which converted to FIFO via `queue.shift()` plus a
regression test. The audit was performed against an earlier PR head.

Tests: 152 packages/api + 195 api JS = 347 pass. Typecheck clean.

* 🪛 fix: Pure-Subagent codeEnv + Primed-Skill Routing + ToolService Early Returns

Three findings from the second-pass review:

**P2 — Pure subagents missed `codeEnvAvailable`** (initialize.js).
The pure-subagent init path didn't forward the endpoint-level
`codeEnvAvailable` flag to `initializeAgent`, unlike the primary,
handoff, and addedConvo paths. A code-enabled subagent loaded only
through `subagentAgentConfigs` initialized with
`codeEnvAvailable: false`, so even though the recursive seed walk
found its primed code files, the subagent's own `bash_tool` /
`read_file` sandbox fallback were silently gated off. Forward the
flag and add `codeEnvAvailable: config.codeEnvAvailable` to the
`agentToolContexts.set` for symmetry with the other paths.

**P2 — Primed skills outside the catalog cap were misrouted to
sandbox** (handlers.ts). Manual ($-popover) and always-apply primes
are intentionally resolved off the wider `accessibleSkillIds` ACL
set BEFORE catalog injection — see `resolveManualSkills` for why a
skill outside the `SKILL_CATALOG_LIMIT` cap can still be authorized
for direct manual invocation. The `activeSkillNames` shortcut ran
before reading `skillPrimedIdsByName`, so a primed skill not in the
catalog would fall through to the sandbox instead of resolving via
the pinned `_id`. Read the primed map first and bypass the shortcut
for primed names. New regression test asserts a primed-but-not-
cataloged skill resolves through the existing skill path with
`getSkillByName` invoked and `readSandboxFile` NOT called.

**P3 — `loadAgentTools` early returns dropped `primedCodeFiles`**
(ToolService.js). The non-`definitionsOnly` path captures the field
correctly, but two early-return branches (no-action-tools fast path,
no-action-sets fast path) omitted it. Any traditional
`loadAgentTools(..., definitionsOnly: false)` caller using
execute_code without action tools would have its first-call session
seed silently empty. Add `primedCodeFiles` to both early returns
for consistency with the final return shape.

Tests: 153 packages/api + 195 api JS = 348 pass.

* 🧹 chore: Document jest.mock arrow-indirection pattern in process.spec.js

Per the second-pass review's Finding #2 (NIT, "would help future
readers"): the mock setup mixes direct `jest.fn()` references with
arrow-function indirection (`(...args) => mockX(...args)`). The
indirection isn't stylistic — it's required because `jest.mock(...)`
is hoisted above the outer `const` declarations at parse time, so a
direct reference would capture `undefined`. Inline comment explains
the pattern so the next reader doesn't have to reverse-engineer it
or accidentally "simplify" the mocks and break per-test
`mockReturnValueOnce` / `mockImplementationOnce` overrides.

* 🪛 fix: Five Issues from Pass-N + Codex Review (incl. 404 root cause)

Five real bugs surfaced by another review pass + Codex PR comments
+ the codeapi-side logs we collected during manual testing:

**1) `processCodeOutput` 404 root cause (`callbacks.js`).**
The codeapi worker emits TWO distinct `session_id`s on a tool result:
  - `artifact.session_id` is the EXEC session — the sandbox VM that
    ran the bash command. Files don't live there; it's torn down
    post-execution.
  - `file.session_id` is the STORAGE session — the file-server
    bucket prefix where artifacts actually live.
`callbacks.js` was passing the EXEC id to `processCodeOutput`, which
builds `/download/{session_id}/{id}` and 404s because the file-server
doesn't know about that path. This explains every "Error
downloading/processing code environment file" we saw during testing.
Use `file.session_id ?? output.artifact.session_id` (per-file id with
artifact-level fallback for older worker payloads).

**2) `primeFiles` reupload pushed STALE sandbox ids (`process.js`).**
When `getSessionInfo` returns null (file expired/missing in sandbox),
`reuploadFile` re-uploads via `handleFileUpload`, gets a NEW
`fileIdentifier`, and persists it on the DB record. But `pushFile`
was a closure capturing the OLD `(session_id, id)` parsed at the top
of the loop, so the in-memory `files[]` array (now consumed by
`buildInitialToolSessions` to seed `Graph.sessions`) silently
referenced a sandbox object that no longer existed. The first tool
call would 404 trying to mount it; only the next turn's metadata
re-read would correct course. Parameterize `pushFile` with optional
`(session_id, id)` overrides; in `reuploadFile` parse the new
identifier and pass through. 2 regression tests.

**3) Codex P2 — Cap sandbox fallback output before line-numbering
(`handlers.ts`).** The new `handleSandboxFileFallback` returned
`addLineNumbers(result.content)` without a size guard, so reading a
multi-MB `/mnt/data/*` artifact materialized the file twice in
memory (raw + line-numbered) before downstream truncation. Match the
existing skill-file path's `MAX_READABLE_BYTES` (256KB): truncate
raw first, then number, surface the truncation to the model so it
can use `bash_tool` (`head` / `tail`) for the rest. 2 tests
(oversized truncates with hint, in-cap doesn't).

**4) Codex P2 — Dedupe seeded code files by `(session_id, id)`
(`codeFilesSession.ts`).** Multiple agents in a run commonly carry
the same primed execute-code resources (shared conversation files);
without dedupe, `_injected_files` grows proportionally to agent
count and bloats every `/exec` POST. Use a `(session_id, id)`
identity key so first-seen wins (preserves source ordering); name
alone isn't sufficient because two distinct primed uploads can
share a filename across different sessions. 4 tests covering dedup
across iterations, against pre-existing entries, name-collision
distinct-session preservation, and the multi-agent realistic case
in `buildInitialToolSessions`.

**5) Pass-N P2 — Polyfill `globalThis.File` in api Jest setup
(`api/test/jestSetup.js`).** `packages/api/jest.setup.cjs` had the
polyfill; the legacy api workspace's Jest config has its own
`setupFiles` that didn't, so on Node 18 / WSL the api focused tests
still failed at import time with `ReferenceError: File is not
defined` from undici. Mirror the polyfill.

Tests: 159 packages/api + 206 api JS = 365 pass. Typecheck clean.

* 🔧 chore: Update @librechat/agents dependency to version 3.1.73

Bumps the version of the @librechat/agents package across package-lock.json and relevant package.json files to ensure compatibility with the latest features and fixes.
2026-04-27 08:56:39 +09:00

809 lines
29 KiB
JavaScript

const { logger } = require('@librechat/data-schemas');
const { createContentAggregator } = require('@librechat/agents');
const {
loadSkillStates,
initializeAgent,
primeInvokedSkills,
validateAgentModel,
extractManualSkills,
GenerationJobManager,
getCustomEndpointConfig,
discoverConnectedAgents,
resolveAgentScopedSkillIds,
} = require('@librechat/api');
const {
ResourceType,
EModelEndpoint,
PermissionBits,
isAgentsEndpoint,
getResponseSender,
AgentCapabilities,
isEphemeralAgentId,
} = require('librechat-data-provider');
const {
createToolEndCallback,
getDefaultHandlers,
} = require('~/server/controllers/agents/callbacks');
const { loadAgentTools, loadToolsForExecution } = require('~/server/services/ToolService');
const { filterFilesByAgentAccess } = require('~/server/services/Files/permissions');
const {
getSkillToolDeps,
enrichWithSkillConfigurable,
buildSkillPrimedIdsByName,
} = require('./skillDeps');
const { getModelsConfig } = require('~/server/controllers/ModelController');
const { checkPermission, findAccessibleResources } = require('~/server/services/PermissionService');
const AgentClient = require('~/server/controllers/agents/client');
const { processAddedConvo } = require('./addedConvo');
const { logViolation } = require('~/cache');
const db = require('~/models');
/**
* Creates a tool loader function for the agent.
* @param {AbortSignal} signal - The abort signal
* @param {string | null} [streamId] - The stream ID for resumable mode
* @param {boolean} [definitionsOnly=false] - When true, returns only serializable
* tool definitions without creating full tool instances (for event-driven mode)
*/
function createToolLoader(signal, streamId = null, definitionsOnly = false) {
/**
* @param {object} params
* @param {ServerRequest} params.req
* @param {ServerResponse} params.res
* @param {string} params.agentId
* @param {string[]} params.tools
* @param {string} params.provider
* @param {string} params.model
* @param {AgentToolResources} params.tool_resources
* @returns {Promise<{
* tools?: StructuredTool[],
* toolContextMap: Record<string, unknown>,
* toolDefinitions?: import('@librechat/agents').LCTool[],
* userMCPAuthMap?: Record<string, Record<string, string>>,
* toolRegistry?: import('@librechat/agents').LCToolRegistry
* } | undefined>}
*/
return async function loadTools({
req,
res,
tools,
model,
agentId,
provider,
tool_options,
tool_resources,
}) {
const agent = { id: agentId, tools, provider, model, tool_options };
try {
return await loadAgentTools({
req,
res,
agent,
signal,
streamId,
tool_resources,
definitionsOnly,
});
} catch (error) {
logger.error('Error loading tools for agent ' + agentId, error);
}
};
}
/**
* Initializes the AgentClient for a given request/response cycle.
* @param {Object} params
* @param {Express.Request} params.req
* @param {Express.Response} params.res
* @param {AbortSignal} params.signal
* @param {Object} params.endpointOption
*/
const initializeClient = async ({ req, res, signal, endpointOption }) => {
if (!endpointOption) {
throw new Error('Endpoint option not provided');
}
const appConfig = req.config;
/** @type {string | null} */
const streamId = req._resumableStreamId || null;
/** @type {Array<UsageMetadata>} */
const collectedUsage = [];
/** @type {ArtifactPromises} */
const artifactPromises = [];
const { contentParts, aggregateContent } = createContentAggregator();
const toolEndCallback = createToolEndCallback({ req, res, artifactPromises, streamId });
/** Query accessible skill IDs once per run (shared across all agents).
* Skills activate under strict opt-in semantics — see
* `resolveAgentScopedSkillIds` for the per-agent activation predicate:
* - Ephemeral agent → per-conversation skills badge toggle (full catalog).
* - Persisted agent → `agent.skills_enabled === true`. Optional
* `agent.skills` allowlist narrows the catalog; empty/undefined
* allowlist with the toggle on = full accessible catalog. */
const enabledCapabilities = new Set(appConfig?.endpoints?.[EModelEndpoint.agents]?.capabilities);
const skillsCapabilityEnabled = enabledCapabilities.has(AgentCapabilities.skills);
const codeEnvAvailable = enabledCapabilities.has(AgentCapabilities.execute_code);
const ephemeralSkillsToggle = req.body?.ephemeralAgent?.skills === true;
const accessibleSkillIds = skillsCapabilityEnabled
? await findAccessibleResources({
userId: req.user.id,
role: req.user.role,
resourceType: ResourceType.SKILL,
requiredPermissions: PermissionBits.VIEW,
})
: [];
const { skillStates, defaultActiveOnShare } = await loadSkillStates({
userId: req.user.id,
appConfig,
getUserById: db.getUserById,
accessibleSkillIds,
});
/**
* Agent context store - populated after initialization, accessed by callback via closure.
* Maps agentId -> { userMCPAuthMap, agent, tool_resources, toolRegistry, openAIApiKey }
* @type {Map<string, {
* userMCPAuthMap?: Record<string, Record<string, string>>,
* agent?: object,
* tool_resources?: object,
* toolRegistry?: import('@librechat/agents').LCToolRegistry,
* openAIApiKey?: string
* }>}
*/
const agentToolContexts = new Map();
const toolExecuteOptions = {
loadTools: async (toolNames, agentId) => {
const ctx = agentToolContexts.get(agentId) ?? {};
logger.debug(`[ON_TOOL_EXECUTE] ctx found: ${!!ctx.userMCPAuthMap}, agent: ${ctx.agent?.id}`);
logger.debug(`[ON_TOOL_EXECUTE] toolRegistry size: ${ctx.toolRegistry?.size ?? 'undefined'}`);
const result = await loadToolsForExecution({
req,
res,
signal,
streamId,
toolNames,
agent: ctx.agent,
toolRegistry: ctx.toolRegistry,
userMCPAuthMap: ctx.userMCPAuthMap,
tool_resources: ctx.tool_resources,
actionsEnabled: ctx.actionsEnabled,
});
logger.debug(`[ON_TOOL_EXECUTE] loaded ${result.loadedTools?.length ?? 0} tools`);
/** Per-agent narrowed flag (admin capability AND agent.tools
* includes execute_code), captured in `agentToolContexts` when
* the agent initialized. Falls back to `false` on any stray
* ctx miss so a skills-only agent never gains sandbox access
* even if capability lookup somehow skips. */
return enrichWithSkillConfigurable(
result,
req,
ctx.accessibleSkillIds,
ctx.codeEnvAvailable === true,
ctx.skillPrimedIdsByName,
ctx.activeSkillNames,
);
},
toolEndCallback,
...getSkillToolDeps(),
};
const summarizationOptions =
appConfig?.summarization?.enabled === false ? { enabled: false } : { enabled: true };
/**
* Per-request map of per-subagent `createContentAggregator` instances
* keyed by the parent's `tool_call_id`. The handler in `callbacks.js`
* lazily creates an aggregator for each distinct `parentToolCallId`
* and folds every `ON_SUBAGENT_UPDATE` event into it as they stream
* in. `AgentClient` pulls each aggregator's `contentParts` at message
* save time and attaches them to the matching `subagent` tool_call so
* the child's reasoning / tool calls / final text survive a page
* refresh — the client-side Recoil atom is best-effort live-only.
*/
const subagentAggregatorsByToolCallId = new Map();
const eventHandlers = getDefaultHandlers({
res,
toolExecuteOptions,
summarizationOptions,
aggregateContent,
toolEndCallback,
collectedUsage,
streamId,
subagentAggregatorsByToolCallId,
});
if (!endpointOption.agent) {
throw new Error('No agent promise provided');
}
const primaryAgent = await endpointOption.agent;
delete endpointOption.agent;
if (!primaryAgent) {
throw new Error('Agent not found');
}
const modelsConfig = await getModelsConfig(req);
const validationResult = await validateAgentModel({
req,
res,
modelsConfig,
logViolation,
agent: primaryAgent,
});
if (!validationResult.isValid) {
throw new Error(validationResult.error?.message);
}
const agentConfigs = new Map();
const allowedProviders = new Set(appConfig?.endpoints?.[EModelEndpoint.agents]?.allowedProviders);
/** Event-driven mode: only load tool definitions, not full instances */
const loadTools = createToolLoader(signal, streamId, true);
/** @type {Array<MongoFile>} */
const requestFiles = req.body.files ?? [];
/** @type {string} */
const conversationId = req.body.conversationId;
/** @type {string | undefined} */
const parentMessageId = req.body.parentMessageId;
/**
* Skill names the user invoked via the `$` popover for this turn. Only flows
* to the primary agent — handoff agents are follow-up turns that don't see
* the user's per-submission `$` selections. `extractManualSkills` also
* drops non-string / empty elements so a crafted payload can't reach the
* `getSkillByName` DB query with nonsense values.
* @type {string[] | undefined}
*/
const manualSkills = extractManualSkills(req.body);
const primaryScopedSkillIds = resolveAgentScopedSkillIds({
agent: primaryAgent,
accessibleSkillIds,
skillsCapabilityEnabled,
ephemeralSkillsToggle,
});
const primaryConfig = await initializeAgent(
{
req,
res,
loadTools,
requestFiles,
conversationId,
parentMessageId,
agent: primaryAgent,
endpointOption,
allowedProviders,
isInitialAgent: true,
accessibleSkillIds: primaryScopedSkillIds,
codeEnvAvailable,
skillStates,
defaultActiveOnShare,
manualSkills,
},
{
getFiles: db.getFiles,
getUserKey: db.getUserKey,
getMessages: db.getMessages,
getConvoFiles: db.getConvoFiles,
updateFilesUsage: db.updateFilesUsage,
getUserKeyValues: db.getUserKeyValues,
getUserCodeFiles: db.getUserCodeFiles,
getToolFilesByIds: db.getToolFilesByIds,
getCodeGeneratedFiles: db.getCodeGeneratedFiles,
filterFilesByAgentAccess,
listSkillsByAccess: db.listSkillsByAccess,
listAlwaysApplySkills: db.listAlwaysApplySkills,
getSkillByName: db.getSkillByName,
},
);
logger.debug(
`[initializeClient] Storing tool context for ${primaryConfig.id}: ${primaryConfig.toolDefinitions?.length ?? 0} tools, registry size: ${primaryConfig.toolRegistry?.size ?? '0'}`,
);
/** Maps each primed skill name (manual `$` or always-apply) to the
* `_id` of the exact doc that was primed. Plumbed to
* `enrichWithSkillConfigurable` so the read_file handler can pin
* same-name collision lookups to the resolver's chosen doc AND relax
* the disable-model-invocation gate for skills whose body is already
* in this turn's context. */
const skillPrimedIdsByName = buildSkillPrimedIdsByName(
primaryConfig.manualSkillPrimes,
primaryConfig.alwaysApplySkillPrimes,
);
agentToolContexts.set(primaryConfig.id, {
agent: primaryAgent,
toolRegistry: primaryConfig.toolRegistry,
userMCPAuthMap: primaryConfig.userMCPAuthMap,
tool_resources: primaryConfig.tool_resources,
actionsEnabled: primaryConfig.actionsEnabled,
accessibleSkillIds: primaryConfig.accessibleSkillIds,
activeSkillNames: primaryConfig.activeSkillNames,
codeEnvAvailable: primaryConfig.codeEnvAvailable,
skillPrimedIdsByName,
});
const {
agentConfigs: discoveredConfigs,
edges: discoveredEdges,
userMCPAuthMap: discoveredMCPAuthMap,
skippedAgentIds: discoveredSkippedIds,
} = await discoverConnectedAgents(
{
req,
res,
primaryConfig,
agent_ids: primaryConfig.agent_ids,
endpointOption,
allowedProviders,
modelsConfig,
loadTools,
requestFiles,
conversationId,
parentMessageId,
computeAccessibleSkillIds: (agent) =>
resolveAgentScopedSkillIds({
agent,
accessibleSkillIds,
skillsCapabilityEnabled,
ephemeralSkillsToggle,
}),
skillStates,
defaultActiveOnShare,
codeEnvAvailable,
},
{
getAgent: db.getAgent,
checkPermission,
logViolation,
db: {
getFiles: db.getFiles,
getUserKey: db.getUserKey,
getMessages: db.getMessages,
getConvoFiles: db.getConvoFiles,
updateFilesUsage: db.updateFilesUsage,
getUserKeyValues: db.getUserKeyValues,
getUserCodeFiles: db.getUserCodeFiles,
getToolFilesByIds: db.getToolFilesByIds,
getCodeGeneratedFiles: db.getCodeGeneratedFiles,
filterFilesByAgentAccess,
listSkillsByAccess: db.listSkillsByAccess,
listAlwaysApplySkills: db.listAlwaysApplySkills,
getSkillByName: db.getSkillByName,
},
// The callback fires during BFS, before the helper prunes agents
// whose edges end up filtered. Don't populate `agentConfigs` here —
// `discoveredConfigs` (returned below) is the authoritative pruned
// set. The per-agent tool context map is OK to keep populated even
// for pruned ids: it's only read by closure in ON_TOOL_EXECUTE,
// stale entries are unreachable at runtime.
//
// Handoff agents get the same `skillPrimedIdsByName` plumbing as the
// primary so `read_file` can pin same-name collisions to the exact
// primed doc AND relax the `disable-model-invocation: true` gate for
// skills whose body is already in this turn's context — matters for
// handoff agents that have their own always-apply skills bound or
// that the user `$`-invokes within the handoff flow.
onAgentInitialized: (agentId, agent, config) => {
agentToolContexts.set(agentId, {
agent,
toolRegistry: config.toolRegistry,
userMCPAuthMap: config.userMCPAuthMap,
tool_resources: config.tool_resources,
actionsEnabled: config.actionsEnabled,
accessibleSkillIds: config.accessibleSkillIds,
activeSkillNames: config.activeSkillNames,
codeEnvAvailable: config.codeEnvAvailable,
skillPrimedIdsByName: buildSkillPrimedIdsByName(
config.manualSkillPrimes,
config.alwaysApplySkillPrimes,
),
});
},
// Pass through the `@librechat/api` exports so that tests which
// `jest.mock('@librechat/api')` can override the initializer/validator.
initializeAgent,
validateAgentModel,
},
);
// Copy the pruned discovery result into the outer map. Anything the
// helper dropped (skipped or unreachable after edge filtering) is
// intentionally absent. `processAddedConvo` below may still add more
// entries for parallel multi-convo execution.
for (const [agentId, config] of discoveredConfigs) {
agentConfigs.set(agentId, config);
}
let userMCPAuthMap = discoveredMCPAuthMap;
let edges = discoveredEdges;
/** Multi-Convo: Process addedConvo for parallel agent execution */
const { userMCPAuthMap: updatedMCPAuthMap } = await processAddedConvo({
req,
res,
loadTools,
logViolation,
modelsConfig,
requestFiles,
agentConfigs,
primaryAgent,
endpointOption,
userMCPAuthMap,
conversationId,
parentMessageId,
allowedProviders,
primaryAgentId: primaryConfig.id,
codeEnvAvailable,
});
if (updatedMCPAuthMap) {
userMCPAuthMap = updatedMCPAuthMap;
}
for (const [agentId, config] of agentConfigs) {
if (agentToolContexts.has(agentId)) {
continue;
}
agentToolContexts.set(agentId, {
agent: config,
toolRegistry: config.toolRegistry,
userMCPAuthMap: config.userMCPAuthMap,
tool_resources: config.tool_resources,
actionsEnabled: config.actionsEnabled,
accessibleSkillIds: config.accessibleSkillIds,
activeSkillNames: config.activeSkillNames,
codeEnvAvailable: config.codeEnvAvailable,
});
}
// `discoverConnectedAgents` always returns a concrete array, so no
// further normalization is needed before handing this to `createRun`.
primaryConfig.edges = edges;
// Subagents: load any explicit subagent configs. Subagents run in isolated
// context windows and are invoked via a dedicated spawn tool (not handoff
// edges). An agent that is ONLY referenced as a subagent is dropped from
// `agentConfigs` so the LangGraph pipeline doesn't treat it as a
// parallel/handoff node, but it is KEPT in `agentToolContexts` — the child's
// `ON_TOOL_EXECUTE` dispatches resolve tool execution context (agent,
// tool_resources, skill ACLs, ...) from that map, so removing it would leave
// action tools skipped and resource-scoped tools running without their
// configured resources.
const subagentsCapabilityEnabled = enabledCapabilities.has(AgentCapabilities.subagents);
/** Track skipped ids locally so repeated failures short-circuit within
* the subagent loading loop. Seeded from the discovery helper's skip
* list so agents that already failed handoff loading don't get retried. */
const skippedAgentIds = new Set(discoveredSkippedIds ?? []);
/** All agent ids referenced on any edge (source OR target). Used by
* `loadSubagentsFor` to decide whether an agent that's only a subagent
* can be safely dropped from `agentConfigs` — LangGraph doesn't treat
* pure subagents as parallel/handoff nodes. */
const edgeAgentIds = new Set([primaryConfig.id]);
for (const edge of edges ?? []) {
const sources = Array.isArray(edge.from) ? edge.from : [edge.from];
const targets = Array.isArray(edge.to) ? edge.to : [edge.to];
for (const id of sources) {
if (typeof id === 'string') edgeAgentIds.add(id);
}
for (const id of targets) {
if (typeof id === 'string') edgeAgentIds.add(id);
}
}
/** Lazy per-id agent loader used for subagents that weren't reachable
* via the handoff edge graph (so `discoverConnectedAgents` didn't
* initialize them). Mirrors the helper's internal `processAgent`:
* DB lookup + VIEW check + `initializeAgent`, then inserts into
* `agentConfigs` and `agentToolContexts`. Returns `null` on any
* failure so the caller can skip gracefully. */
const loadAgentById = async (agentId) => {
if (skippedAgentIds.has(agentId)) return null;
const existing = agentConfigs.get(agentId);
if (existing) return existing;
try {
const agent = await db.getAgent({ id: agentId });
if (!agent) {
skippedAgentIds.add(agentId);
return null;
}
const userId = req.user?.id;
if (!userId) {
skippedAgentIds.add(agentId);
return null;
}
const hasAccess = await checkPermission({
userId,
role: req.user?.role,
resourceType: ResourceType.AGENT,
resourceId: agent._id,
requiredPermission: PermissionBits.VIEW,
});
if (!hasAccess) {
logger.warn(
`[processAgent] User ${userId} lacks VIEW access to subagent ${agentId}, skipping`,
);
skippedAgentIds.add(agentId);
return null;
}
const validation = await validateAgentModel({
req,
res,
agent,
modelsConfig,
logViolation,
});
if (!validation.isValid) {
logger.warn(
`[processAgent] Subagent ${agentId} failed model validation: ${validation.error?.message}`,
);
skippedAgentIds.add(agentId);
return null;
}
const config = await initializeAgent(
{
req,
res,
agent,
loadTools,
requestFiles,
conversationId,
parentMessageId,
endpointOption: { ...endpointOption, endpoint: EModelEndpoint.agents },
allowedProviders,
accessibleSkillIds: resolveAgentScopedSkillIds({
agent,
accessibleSkillIds,
skillsCapabilityEnabled,
ephemeralSkillsToggle,
}),
/** Match the primary / handoff / addedConvo paths: forward the
* endpoint-level admin flag so `initializeAgent` can compute the
* per-agent narrowing (admin AND agent.tools includes
* execute_code) into `InitializedAgent.codeEnvAvailable`. Without
* this, a code-enabled subagent loaded only through
* `subagentAgentConfigs` initializes with `codeEnvAvailable:
* false`, so `bash_tool` / `read_file` sandbox fallback are
* silently gated off even though the seed walk found it. */
codeEnvAvailable,
skillStates,
defaultActiveOnShare,
},
{
getAgent: db.getAgent,
checkPermission,
logViolation,
db: {
getFiles: db.getFiles,
getUserKey: db.getUserKey,
getMessages: db.getMessages,
getConvoFiles: db.getConvoFiles,
updateFilesUsage: db.updateFilesUsage,
getUserKeyValues: db.getUserKeyValues,
getUserCodeFiles: db.getUserCodeFiles,
getToolFilesByIds: db.getToolFilesByIds,
getCodeGeneratedFiles: db.getCodeGeneratedFiles,
filterFilesByAgentAccess,
listSkillsByAccess: db.listSkillsByAccess,
listAlwaysApplySkills: db.listAlwaysApplySkills,
getSkillByName: db.getSkillByName,
},
},
);
agentConfigs.set(agentId, config);
agentToolContexts.set(agentId, {
agent,
toolRegistry: config.toolRegistry,
userMCPAuthMap: config.userMCPAuthMap,
tool_resources: config.tool_resources,
actionsEnabled: config.actionsEnabled,
accessibleSkillIds: config.accessibleSkillIds,
activeSkillNames: config.activeSkillNames,
codeEnvAvailable: config.codeEnvAvailable,
skillPrimedIdsByName: buildSkillPrimedIdsByName(
config.manualSkillPrimes,
config.alwaysApplySkillPrimes,
),
});
return config;
} catch (err) {
logger.error(`[processAgent] Error processing subagent ${agentId}:`, err);
skippedAgentIds.add(agentId);
return null;
}
};
/** Collected during resolution; applied to `agentConfigs` only after
* every config has had its subagents resolved. Eager pruning would
* hide pure-subagent ids from the subsequent `loadSubagentsFor`
* loop, which would leave *their* `subagentAgentConfigs` empty and
* silently break nested delegation like A → B → C where B is only
* a subagent of A. */
const pureSubagentIds = new Set();
/**
* Loads `subagentAgentConfigs` for a single agent config. Shared
* between the primary agent and handoff-target agents (and pure
* subagents, transitively) so an agent used via handoff or
* nested-subagent that has its own explicit `subagents.agent_ids`
* gets them honored at runtime. Self-spawn works regardless (no DB
* lookup needed). Pruning decisions are deferred to `pureSubagentIds`.
*/
const loadSubagentsFor = async (config) => {
const sub = config.subagents;
if (!subagentsCapabilityEnabled || !sub?.enabled) {
config.subagentAgentConfigs = [];
return;
}
/** Dedupe and filter in one pass — a crafted payload could
* legitimately include the same ID twice; the backend shouldn't
* create duplicate SubagentConfig entries for the LLM to see as
* separate spawn targets. */
const explicitSubagentIds = Array.from(
new Set(
Array.isArray(sub.agent_ids)
? sub.agent_ids.filter((id) => typeof id === 'string' && id && id !== config.id)
: [],
),
);
/** @type {Array<Object>} */
const resolved = [];
for (const subagentId of explicitSubagentIds) {
if (skippedAgentIds.has(subagentId)) continue;
/** Cycle guard: a configuration like A ↔ B (B lists A as its
* subagent) would otherwise trigger `loadAgentById` on the
* primary — inserting a second config for the same primary id,
* which downstream duplicates in the agent array. Reuse the
* existing primary config when a subagent ref points back at it. */
if (subagentId === primaryConfig.id) {
resolved.push(primaryConfig);
continue;
}
const subagentConfig = await loadAgentById(subagentId);
if (!subagentConfig) continue;
resolved.push(subagentConfig);
if (!edgeAgentIds.has(subagentId)) {
pureSubagentIds.add(subagentId);
}
}
config.subagentAgentConfigs = resolved;
};
/** BFS across the primary's subagent tree so nested chains like
* A → B → C get resolved before any pruning. Each config is
* visited once. */
const visitedConfigIds = new Set();
const pending = [primaryConfig];
while (pending.length > 0) {
const cfg = pending.shift();
if (!cfg || visitedConfigIds.has(cfg.id)) continue;
visitedConfigIds.add(cfg.id);
await loadSubagentsFor(cfg);
for (const child of cfg.subagentAgentConfigs ?? []) {
if (child?.id && !visitedConfigIds.has(child.id)) {
pending.push(child);
}
}
}
/** Handoff targets still in the map that weren't visited via the
* primary's subagent tree also need their subagents resolved. */
for (const [id, cfg] of agentConfigs.entries()) {
if (id === primaryConfig.id || visitedConfigIds.has(id)) continue;
visitedConfigIds.add(id);
await loadSubagentsFor(cfg);
for (const child of cfg.subagentAgentConfigs ?? []) {
if (child?.id && !visitedConfigIds.has(child.id)) {
visitedConfigIds.add(child.id);
await loadSubagentsFor(child);
}
}
}
/** Drop pure-subagent entries now that every reachable config has
* had its subagents resolved. They stay in `agentToolContexts` so
* their tools still execute with the right scoping. */
for (const id of pureSubagentIds) {
agentConfigs.delete(id);
}
primaryConfig.subagents = subagentsCapabilityEnabled ? primaryConfig.subagents : undefined;
/** If the capability is off at the endpoint level, strip `subagents` on
* every loaded config — not just the primary. `run.ts` calls
* `buildSubagentConfigs` for every agent in the array, so a handoff
* agent with `subagents.enabled: true` persisted on its document would
* otherwise still expose self-spawn at runtime even though the admin
* has disabled the capability globally. */
if (!subagentsCapabilityEnabled) {
for (const config of agentConfigs.values()) {
config.subagents = undefined;
config.subagentAgentConfigs = undefined;
}
}
let endpointConfig = appConfig.endpoints?.[primaryConfig.endpoint];
if (!isAgentsEndpoint(primaryConfig.endpoint) && !endpointConfig) {
try {
endpointConfig = getCustomEndpointConfig({
endpoint: primaryConfig.endpoint,
appConfig,
});
} catch (err) {
logger.error(
'[api/server/controllers/agents/client.js #titleConvo] Error getting custom endpoint config',
err,
);
}
}
const sender =
primaryAgent.name ??
getResponseSender({
...endpointOption,
model: endpointOption.model_parameters.model,
modelDisplayLabel: endpointConfig?.modelDisplayLabel,
modelLabel: endpointOption.model_parameters.modelLabel,
});
/** History priming uses the user's full ACL-accessible skill set (not
* per-agent scoped) because prior turns may reference skills no longer
* in any active agent's scope; the ACL check is the security gate.
* `codeEnvAvailable` comes from `primaryConfig` — @see
* `InitializedAgent.codeEnvAvailable` for the per-agent narrowing. */
const handlePrimeInvokedSkills = skillsCapabilityEnabled
? (payload) =>
primeInvokedSkills({
req,
payload,
accessibleSkillIds,
codeEnvAvailable: primaryConfig.codeEnvAvailable === true,
...getSkillToolDeps(),
})
: undefined;
const client = new AgentClient({
req,
res,
sender,
contentParts,
agentConfigs,
eventHandlers,
collectedUsage,
aggregateContent,
artifactPromises,
primeInvokedSkills: handlePrimeInvokedSkills,
agent: primaryConfig,
spec: endpointOption.spec,
iconURL: endpointOption.iconURL,
attachments: primaryConfig.attachments,
endpointType: endpointOption.endpointType,
resendFiles: primaryConfig.resendFiles ?? true,
maxContextTokens: primaryConfig.maxContextTokens,
endpoint: isEphemeralAgentId(primaryConfig.id) ? primaryConfig.endpoint : EModelEndpoint.agents,
subagentAggregatorsByToolCallId,
});
if (streamId) {
GenerationJobManager.setCollectedUsage(streamId, collectedUsage);
}
return { client, userMCPAuthMap };
};
module.exports = { initializeClient };