Skip to main content
Glama

codex_request

Destructive

Run OpenAI Codex CLI tasks by submitting a prompt and receiving a synchronous result, or a pollable job when execution exceeds the deadline.

Instructions

Run an OpenAI Codex CLI request synchronously (when async jobs are enabled, auto-defers to a pollable job past the sync deadline; otherwise runs to completion). Requires exactly one of prompt or promptParts.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
ossNoCodex --oss: use the open-source provider. New sessions only.
colorNoCodex --color: output color mode (always|never|auto). New sessions only.
modelNoModel name or alias (e.g. gpt-5.5, gpt-5.4, latest)
addDirNoCodex --add-dir <DIR>: additional writable workspace directories. Emitted once per entry on new sessions only: the flag is filtered from resume argv. That does NOT pin a resumed session's writable-dir policy; do not rely on it. Stdio/local callers may pass local paths directly. Remote HTTP/OAuth callers must use relative paths inside a selected registered workspace. Do not call workspace_* tools to fix stdio/local provider path access.
enableNoCodex --enable <FEATURE> (repeatable): enable a feature for this run (equivalent to -c features.<name>=true). Accepted on resume. Local callers only, remote HTTP/OAuth requests are rejected.
imagesNoCodex -i <path>: image attachments. Each path must exist; missing paths fail fast.
promptNoPrompt text for Codex (mutually exclusive with promptParts)
searchNoDEPRECATED compatibility input: accepted but ignored because current Codex exec no longer accepts --search.
disableNoCodex --disable <FEATURE> (repeatable): disable a feature for this run (equivalent to -c features.<name>=false). Accepted on resume. Local callers only, remote HTTP/OAuth requests are rejected.
profileNoCodex --profile <name>: select a profile from ~/.codex/config.toml.
fullAutoNoDEPRECATED: prefer `sandboxMode`. Expands to `--sandbox workspace-write`; current Codex no longer accepts approval-policy flags.
worktreeNoSlice λ: run this request inside a dedicated git worktree owned by the gateway. `true` creates a fresh worktree at `<repoRoot>/.worktrees/<uuid>` branched from HEAD. `{ name?, ref? }` lets the caller supply a sanitized name and/or a git ref (default: HEAD). When the request carries a sessionId and the session already has a worktree, reuse requires same-host ownership metadata and a matching live Git registration. Named path collisions never reuse manager state. Gateway-managed worktrees work with file-backed and PostgreSQL session managers. Shared PostgreSQL rows do not transfer filesystem ownership; reuse and cleanup remain limited to the owning host. The Grok, Devin, and Mistral adapters require an explicit provider-native sessionId; fresh, createNewSession, and resumeLatest-only worktree requests fail closed because they cannot durably reselect the worktree. The request must select a registered workspace explicitly, through caller-owned session metadata, or through the configured default. Worktrees never fall back to the gateway process cwd and cannot be combined with local workingDir, addDir, or includeDirs paths. The gateway spawns the child CLI with `cwd: <worktree-path>`; no `-w`/`--worktree` flag is ever emitted to the underlying CLI. On worktree materialization, the gateway suppresses repository, system, and global Git hooks and configured clean, smudge, and process checkout filters, sparse checkout, and lazy object fetching. Filter-dependent content such as Git LFS remains in its repository representation instead of executing host commands. On session_delete, session_clear_all or file-backed TTL eviction, both managers hide a durably owned worktree session while cleanup runs. Failed Git removal retains a cleanup-pending tombstone, blocks reuse, and is retried when a manager is registered on the owning host; the record is finalized only once Git no longer registers the worktree, which is read back rather than inferred from the recorded path being absent. A worktree that moved, or whose owner marker cannot be read, is not a removal. Deletion processed by another host removes no worktree and leaves the owning host's record intact. The database-side cleanup_expired_sessions function stages the same tombstone instead of deleting a worktree-bearing session; it invokes no gateway observer and so attempts no removal itself. A tombstone is not bounded by retention. Successful responses are prefixed with `[gateway] worktree=<absolute-path>\n` so callers can use the path. For Claude approvalStrategy:mcp_managed, requesting or reusing a worktree requires approval and LLM_GATEWAY_APPROVAL_ALLOW_BYPASS=1. Other adapters reject mcp_managed before launch. NOTE: callers should `.gitignore` the `.worktrees/` directory in their repo (the gateway does NOT auto-gitignore; see slice λ spec Q4).
ephemeralNoCodex --ephemeral: do not persist the session to disk.
sessionIdNoCodex session UUID to resume via `codex exec resume <ID>`. Must be a real Codex session ID (from `~/.codex/sessions/` or the `codex resume` picker). Gateway-generated `gw-*` IDs are rejected. For a brand-new session no resumable sessionId is returned; continue with resumeLatest:true or a real Codex UUID.
workspaceNoRegistered workspace alias for remote HTTP/OAuth provider calls. Do not use this field, workspace_list, or workspace_register_existing_repo as a fallback for stdio/local provider path access; pass workingDir/addDir/includeDirs directly instead.
mcpServersNoThis provider does not receive gateway-managed MCP configuration. mcpServers has no effect with legacy, and mcp_managed is rejected before launch because ambient MCP configuration cannot be isolated.
workingDirNoCodex -C/--cd <DIR>: working root for this session. Emitted on new sessions only: the flag is filtered from resume argv. That does NOT pin a resumed session's directory, because the child is still spawned with the gateway-resolved cwd; do not rely on a resumed session's working directory. Personal Agent Config Kit mode requires an absolute path. Stdio/local callers may pass local paths directly. Remote HTTP/OAuth callers must use relative paths inside a selected registered workspace. Do not call workspace_* tools to fix stdio/local provider path access.
ignoreRulesNoCodex --ignore-rules: skip project rule files for this run.
promptPartsNoCache-aware structured prompt: { system?, tools?, context?, task }. Mutually exclusive with prompt. Stable parts hash into cache_state for prefix-discipline tracking.
sandboxModeNoCodex --sandbox. NEW SESSIONS ONLY: the gateway filters --sandbox out of a resume argv, so this field has no effect on a resumed request. That is NOT a guarantee that the resumed session keeps its original posture: configOverrides still passes through and can set sandbox_mode, and Codex re-resolves configuration on a cold resume. Establish the posture on the first request and verify it when it matters. On a new session, omitting it does NOT guarantee read-only: the gateway emits no --sandbox flag and Codex resolves the policy from configuration, project trust, and its own fallback, so a trusted project can resolve to workspace-write. Pass read-only explicitly for inspection. Pass workspace-write to let Codex edit files in the working dir, or danger-full-access for unrestricted access.
forceRefreshNoBypass dedup and force a fresh CLI run even if a recent identical request exists
outputFormatNoCodex caller-facing output format. Token/cache usage is recorded in the flight recorder regardless. `text` (default) returns the plain reply; `json` returns the raw `--json` JSONL event stream.
outputSchemaNoCodex --output-schema. Pass a path (string) or an inline JSON Schema object; object is materialised to a 0o600 temp file under os.tmpdir() and deleted after the run.
resumeLatestNoResume a previous Codex session via `codex exec resume --last`. UNDER REVIEW: do not rely on which session this selects or on the resumed working directory. `--last` is filtered by cwd upstream unless `--all` is passed, which the gateway does not emit, and the child is still spawned with the gateway-resolved cwd even though `-C`/`--add-dir` are dropped from the resume argv. Verify the target, or start a fresh session when it must be certain. Ignored if sessionId is set; an explicit real Codex UUID targets that session. A brand-new session returns no resumable sessionId; continue with resumeLatest:true or a real Codex UUID.
strictConfigNoCodex --strict-config: error out when config.toml contains fields not recognized by this Codex version. New sessions only.
correlationIdNoRequest trace ID (auto if omitted)
idleTimeoutMsNoIdle timeout in ms (min 30s, max 1h, omit=CLI default)
localProviderNoCodex --local-provider: local OSS provider (lmstudio|ollama), used with oss. New sessions only.
providerFlagsNoFlags passed to the provider binary verbatim, keyed exactly as the binary spells them (e.g. {"--best-of-n": "3", "--verbatim": true, "--rules": ["a", "b"]}). Use this for any flag your installed CLI accepts that this schema does not name: the binary decides what it supports, not the gateway. true emits the flag alone; a list REPEATS the flag once per item (pass a joined string if your CLI wants a comma-separated value). Values may not start with '-', and a flag the gateway is already emitting for this request is refused rather than duplicated. LOCAL stdio callers only: remote HTTP/OAuth callers are refused every flag here and should use this tool's declared parameters, which carry their own host-path and approval gates.
approvalPolicyNoapprovalPolicy is unavailable for this provider. It has no effect with legacy, and mcp_managed is rejected before launch because ambient MCP configuration cannot be isolated.
askForApprovalNoDEPRECATED compatibility input: accepted but ignored because current Codex no longer accepts --ask-for-approval.
optimizePromptNoOptimize prompt before execution
configOverridesNoCodex -c key=value overrides. Keys: /^[a-zA-Z0-9._]+$/. Values: no CR/LF. Local callers only, remote HTTP/OAuth requests are rejected.
approvalStrategyNoApproval strategy: legacy is supported. mcp_managed is rejected before launch because this adapter cannot isolate ambient MCP configuration.legacy
compressResponseNoCompress the response display text via the native compressor (default: [compression].enabled in config.toml, off unless opted in). Skipped for structured output.
createNewSessionNoForce a fresh session (no resume)
ignoreUserConfigNoCodex --ignore-user-config: ignore ~/.codex/config.toml for this run.
optimizeResponseNoOptimize response output
outputLastMessageNoCodex -o/--output-last-message <FILE>: write the agent's last message to a file. New sessions only.
requestInstructionsNoPer-request Kit instructions. Available only when [personal_config].enabled = true.
useLegacyFullAutoFlagNoDEPRECATED compatibility input: accepted but ignored because current Codex no longer accepts --full-auto.
dangerouslyBypassHookTrustNoCodex --dangerously-bypass-hook-trust: run enabled hooks without persisted hook trust for this invocation. DANGEROUS.
dangerouslyBypassApprovalsAndSandboxNoRun Codex without approvals/sandbox.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changedv3.2.1
    • changedInput schema / properties / worktree / description
      Previous value: -"Slice λ: run this request inside a dedicated git worktree owned by the gateway. `true` creates a fresh worktree at `<repoRoot>/.worktrees/<uuid>` branched from HEAD. `{ name?, ref? }` lets the caller supply a sanitized name and/or a git ref (default: HEAD). When the request carries a sessionId and the session already has a worktree, reuse requires same-host ownership metadata and a matching live Git registration. Named path collisions never reuse manager state. Gateway-managed worktrees require the local file-backed session manager and fail closed with PostgreSQL sessions. The Grok, Devin, and Mistral adapters require an explicit provider-native sessionId; fresh, createNewSession, and resumeLatest-only worktree requests fail closed because they cannot durably reselect the worktree. The request must select a registered workspace explicitly, through caller-owned session metadata, or through the configured default. Worktrees never fall back to the gateway process cwd and cannot be combined with local workingDir, addDir, or includeDirs paths. The gateway spawns the child CLI with `cwd: <worktree-path>`; no `-w`/`--worktree` flag is ever emitted to the underlying CLI. On worktree materialization, the gateway suppresses repository, system, and global Git hooks and configured clean, smudge, and process checkout filters, sparse checkout, and lazy object fetching. Filter-dependent content such as Git LFS remains in its repository representation instead of executing host commands. On session_delete or TTL eviction the gateway hides the session and runs `git worktree remove --force`. Failed removal retains a durable cleanup-pending tombstone which blocks reuse and is retried when the file store is registered on the owning host. The tombstone is finalized only after verified Git removal. Successful responses are prefixed with `[gateway] worktree=<absolute-path>\\n` so callers can use the path. For Claude approvalStrategy:mcp_managed, requesting or reusing a worktree requires approval and LLM_GATEWAY_APPROVAL_ALLOW_BYPASS=1. Other adapters reject mcp_managed before launch. NOTE: callers should `.gitignore` the `.worktrees/` directory in their repo (the gateway does NOT auto-gitignore; see slice λ spec Q4)."New value: +"Slice λ: run this request inside a dedicated git worktree owned by the gateway. `true` creates a fresh worktree at `<repoRoot>/.worktrees/<uuid>` branched from HEAD. `{ name?, ref? }` lets the caller supply a sanitized name and/or a git ref (default: HEAD). When the request carries a sessionId and the session already has a worktree, reuse requires same-host ownership metadata and a matching live Git registration. Named path collisions never reuse manager state. Gateway-managed worktrees work with file-backed and PostgreSQL session managers. Shared PostgreSQL rows do not transfer filesystem ownership; reuse and cleanup remain limited to the owning host. The Grok, Devin, and Mistral adapters require an explicit provider-native sessionId; fresh, createNewSession, and resumeLatest-only worktree requests fail closed because they cannot durably reselect the worktree. The request must select a registered workspace explicitly, through caller-owned session metadata, or through the configured default. Worktrees never fall back to the gateway process cwd and cannot be combined with local workingDir, addDir, or includeDirs paths. The gateway spawns the child CLI with `cwd: <worktree-path>`; no `-w`/`--worktree` flag is ever emitted to the underlying CLI. On worktree materialization, the gateway suppresses repository, system, and global Git hooks and configured clean, smudge, and process checkout filters, sparse checkout, and lazy object fetching. Filter-dependent content such as Git LFS remains in its repository representation instead of executing host commands. On session_delete, session_clear_all or file-backed TTL eviction, both managers hide a durably owned worktree session while cleanup runs. Failed Git removal retains a cleanup-pending tombstone, blocks reuse, and is retried when a manager is registered on the owning host; the record is finalized only once Git no longer registers the worktree, which is read back rather than inferred from the recorded path being absent. A worktree that moved, or whose owner marker cannot be read, is not a removal. Deletion processed by another host removes no worktree and leaves the owning host's record intact. The database-side cleanup_expired_sessions function stages the same tombstone instead of deleting a worktree-bearing session; it invokes no gateway observer and so attempts no removal itself. A tombstone is not bounded by retention. Successful responses are prefixed with `[gateway] worktree=<absolute-path>\\n` so callers can use the path. For Claude approvalStrategy:mcp_managed, requesting or reusing a worktree requires approval and LLM_GATEWAY_APPROVAL_ALLOW_BYPASS=1. Other adapters reject mcp_managed before launch. NOTE: callers should `.gitignore` the `.worktrees/` directory in their repo (the gateway does NOT auto-gitignore; see slice λ spec Q4)."
  2. Addedv3.2.0
  3. Removedv3.0.0
  4. First observedv2.16.0

TDQS

A3.8/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare destructiveHint=true, readOnlyHint=false, and idempotentHint=false, so the safety profile is carried by structured data. The description adds the genuinely new behavioral trait of the sync-vs-deferred execution model and the mutual-exclusion constraint, without contradicting any annotation.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Two sentences, front-loaded with the verb and resource, no filler. The parenthetical is dense but packs the deferral semantics efficiently; the second sentence adds the single most important invocation constraint.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness3/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 43-parameter, destructive, output-schema-less tool this is a thin top-level description, but the per-parameter descriptions are exhaustive (100% coverage) and explain outputFormat, sandbox, and resume semantics. The genuine gaps are what a deferred pollable job handle looks like and how to poll it, plus no stated return-value contract.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the per-parameter docs (sandboxMode, sessionId, worktree, providerFlags, etc.) already carry the semantic burden. The top-level description restates the prompt/promptParts mutual exclusion that the schema already documents, adding little beyond surfacing it prominently.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a specific verb+resource ('Run an OpenAI Codex CLI request') and immediately pins the execution mode ('synchronously'), which distinguishes it from the codex_request_async sibling. The auto-deferral clause clarifies exactly how the sync contract behaves, and the one-of prompt/promptParts constraint names the core invocation requirement.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines3/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The deferral clause ('when async jobs are enabled, auto-defers to a pollable job past the sync deadline; otherwise runs to completion') gives the agent a clear behavioral model of when it gets a direct result versus a job handle. However, it never names the async sibling (codex_request_async) or states when to prefer it, so the when-to-use guidance remains implied rather than explicit.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.