Compiler Enterprise Environment Controls

Use these variables to set organization- or repository-wide defaults and compiler policy without editing individual workflow frontmatter files.

In this enterprise controls reference, OTLP defaults are the scope-sensitive exception: GH_AW_DEFAULT_OTLP_ENDPOINT can be a GitHub Actions variable at repository, organization, or enterprise scope, but the GH_AW_DEFAULT_OTLP_ENDPOINT and GH_AW_DEFAULT_OTLP_HEADERS secrets are limited to repository or organization scope.

Enterprise Control Variables

Variable	Source	Purpose	Applies when
GH_AW_DEFAULT_ARTIFACT_RETENTION_DAYS	GitHub Actions vars.* at runtime	Retention period for all gh-aw artifacts	artifact_retention_days is not set in .github/workflows/aw.json
GH_AW_DEFAULT_MAX_AI_CREDITS	GitHub Actions vars.* at runtime	Default AWF apiProxy.maxAiCredits budget	max-ai-credits is not set in frontmatter or any imported workflow
GH_AW_DEFAULT_MAX_TURN_CACHE_MISSES	Compiler process environment	Default AWF apiProxy.maxCacheMisses guardrail	max-turn-cache-misses is not set in frontmatter or any imported workflow
GH_AW_DEFAULT_DETECTION_MAX_AI_CREDITS	GitHub Actions vars.* at runtime	Default threat-detection AWF apiProxy.maxAiCredits budget	safe-outputs.threat-detection.max-ai-credits is not set
GH_AW_DEFAULT_MAX_DAILY_AI_CREDITS	GitHub Actions vars.* at runtime	Default max-daily-ai-credits guardrail threshold	max-daily-ai-credits is not set in frontmatter or any imported workflow
GH_AW_DEFAULT_ADMISSION_URL	GitHub Actions repository/organization vars.* at runtime	HTTPS /api/admission endpoint; empty disables admission	Emitted when on.admission: true or aw.json "admission": true enables admission jobs
GH_AW_DEFAULT_ADMISSION_AUDIENCE	GitHub Actions repository/organization vars.* at runtime	Admission OIDC audience	Defaults to the effective admission endpoint
GH_AW_DEFAULT_ADMISSION_ON_OUTAGE	GitHub Actions repository/organization vars.* at runtime	Admission outage policy, ignore or block	Defaults to ignore; only transient failures can fail open
GH_AW_DEFAULT_MAX_TURNS	Compiler process environment	Default top-level max-turns	max-turns is not set in frontmatter and the selected engine supports max-turns
GH_AW_DEFAULT_TIMEOUT_MINUTES	GitHub Actions vars.* at runtime	Default agentic_execution step timeout-minutes (20 minutes)	timeout-minutes is not set in frontmatter
GH_AW_DEFAULT_AGENT_JOB_TIMEOUT_MINUTES	GitHub Actions vars.* at runtime	Default generated agent job timeout-minutes (60 minutes)	jobs.agent.timeout-minutes is not set in frontmatter
GH_AW_DEFAULT_DETECTION_JOB_TIMEOUT_MINUTES	GitHub Actions vars.* at runtime	Default generated detection job timeout-minutes (10 minutes)	jobs.detection.timeout-minutes is not set in frontmatter
GH_AW_DEFAULT_DETECTION_MODEL	Compiler process environment	Default threat-detection model	safe-outputs.threat-detection.engine.model is not set
GH_AW_DEFAULT_UTC	Compiler process environment	Default project home UTC offset for rendered CLI timestamps	utc is not set in .github/workflows/aw.json
GH_AW_DEFAULT_MODEL_COPILOT	GitHub Actions vars.* at runtime	Default fallback model for Copilot	GH_AW_MODEL_AGENT_COPILOT / GH_AW_MODEL_DETECTION_COPILOT is unset
GH_AW_DEFAULT_MODEL_CLAUDE	GitHub Actions vars.* at runtime	Default fallback model for Claude	GH_AW_MODEL_AGENT_CLAUDE / GH_AW_MODEL_DETECTION_CLAUDE is unset
GH_AW_DEFAULT_MODEL_CODEX	GitHub Actions vars.* at runtime	Default fallback model for Codex	GH_AW_MODEL_AGENT_CODEX / GH_AW_MODEL_DETECTION_CODEX is unset
GH_AW_DEFAULT_OTLP_ENDPOINT	GitHub Actions secrets.* or vars.* at runtime	Default OTLP exporter endpoint	observability.otlp.endpoint is not set in frontmatter or any imported workflow
GH_AW_DEFAULT_OTLP_HEADERS	GitHub Actions secrets.* at runtime	Default OTLP exporter headers for GH_AW_DEFAULT_OTLP_ENDPOINT	observability.otlp.endpoint is not set in frontmatter or any imported workflow
OTEL_EXPORTER_OTLP_PROTOCOL	GitHub Actions vars.* at runtime	OTLP/HTTP encoding, http/json or http/protobuf	Workflow env does not explicitly define the variable; JavaScript traces-specific protocol takes precedence
GH_AW_OTLP_LOGS_ENABLED	GitHub Actions vars.* at runtime	Opt in to compact unified-session OTLP logs (true); unset or false disables export	The conclusion job has a published unified session and configured OTLP endpoints
GH_AW_OTLP_LOGS_MAX_BYTES	GitHub Actions vars.* at runtime	Maximum serialized OTLP session-log request size in bytes (default 65536)	Session-log export is enabled
GH_AW_REQUIRE_SELF_HOSTED_RUNNERS	Compiler process environment	Fail compilation unless every generated job selects a self-hosted runner	Set to true in a trusted compile environment to enforce self-hosted-only runners

Use gh aw env get and gh aw env update to manage GH_AW_DEFAULT_* and OTEL_EXPORTER_OTLP_PROTOCOL variables in batch. These commands operate on GitHub Actions variables and support repo, org, or enterprise scope, so default_otlp_endpoint can still be managed as an enterprise variable. Set OTLP secrets separately with gh secret set; those secrets are limited to repository or organization scope. The defaults file uses default_-prefixed keys such as default_max_ai_credits, default_max_turn_cache_misses, default_detection_max_ai_credits, default_max_daily_ai_credits, default_timeout_minutes, default_agent_job_timeout_minutes, default_detection_job_timeout_minutes, default_model_copilot, default_otlp_endpoint, default_otlp_protocol, default_artifact_retention_days, and default_utc. For the endpoint, gh aw env update --scope ent writes the GH_AW_DEFAULT_OTLP_ENDPOINT variable, not an endpoint secret. To mask the endpoint value, set GH_AW_DEFAULT_OTLP_ENDPOINT with gh secret set at repository or organization scope instead. GH_AW_DEFAULT_OTLP_HEADERS is always a secret and must also be set with gh secret set at repository or organization scope. If both endpoint values exist, clearing only one leaves the other effective through the fallback expression; clear both the endpoint secret and variable to disable OTLP export.

default_otlp_protocol is the defaults-file key for the standard OTEL_EXPORTER_OTLP_PROTOCOL Actions variable (not a GH_AW_DEFAULT_* variable). It accepts http/json or http/protobuf; null or omission deletes the variable at the selected scope. Recompile existing workflows to add the runtime variable lookup. Explicit workflow env takes precedence. See OpenTelemetry for details.

GH_AW_REQUIRE_SELF_HOSTED_RUNNERS is a compile-time enforcement setting, not a GH_AW_DEFAULT_* default, and is not managed by gh aw env. The same policy can be enabled persistently with require_self_hosted_runners: true in .github/workflows/aw.json, or for a single invocation with gh aw compile --require-self-hosted-runners.

Admission jobs are omitted by default. Enable them with workflow frontmatter on.admission: true or "admission": true in .github/workflows/aw.json, then recompile. Either setting enables both jobs; workflow-level false does not override repository-level true. Admission defaults use default_admission_url, default_admission_audience, and default_admission_on_outage in managed YAML. Provision organization variables in each enrolled organization for enterprise-wide rollout; do not assume an enterprise Actions-variable scope for admission. Repository variables override organization values, so admission is cooperative rather than locked policy. Subsequent variable changes are runtime-only for opted-in workflows.

A valid approval supersedes daily-AIC validation. Its primary and optional detection/evals allowances set the corresponding existing runtime credit limits; omitted optional allowances retain their existing component limits. An empty endpoint skips the admission job and its token acquisition; an ignored outage retains daily validation and existing limits. Denial and malformed responses block activation. Admission identity and credentials stay within the compiler-owned admission job, while activation and downstream jobs consume only validated outputs. The admission and isolated conclusion-admission callback jobs have job-scoped id-token: write, including when the endpoint is empty, because GitHub Actions permissions cannot depend on runtime variables. The ordinary conclusion job receives this permission only when separately configured OTLP OIDC requires it. Activation and agent do not gain this permission for admission. Review existing OIDC relying-party policies accordingly.

The endpoint also enables the isolated conclusion-admission callback job to post a compact unified session and usage artifact IDs to the same origin’s /api/conclusion, after artifact upload and any upload retry have finished. The ordinary conclusion job passes only the compact metadata and artifact IDs through job outputs; it does not receive callback OIDC access unless OTLP OIDC is separately configured. The reusable compact profile excludes agent conversation, tool payloads and unbounded nested data; its JSONL is limited to 128 records/16 KiB and the complete request to 64 KiB. The authority resolves the unchanged full session through authenticated, invocation-bound artifact reads. Failed artifact publication does not suppress compact delivery, but full-session availability remains explicitly unknown without an artifact ID. The callback reconstructs invocation identity and obtains its own fresh OIDC token using the same effective audience. Conclusion delivery is best-effort under both outage policies: failures warn without changing completed workflow results or delaying prior publication attempts. Endpoint enrollment therefore covers compact delivery and protected full-session retrieval as well as admission metadata.

When calling a compiled workflow through workflow_call, add id-token: write to the caller job’s permission envelope and preserve any other permissions required by the reusable workflow. The callee’s admission endpoint is evaluated at runtime, so the caller must grant this scope even when admission is currently disabled:

jobs:
  call-worker:
    uses: ./.github/workflows/worker.lock.yml
    permissions:
      contents: read
      id-token: write

See the admission contract for protocol and deployment requirements.

Required compile check

jobs:
  compile:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
      - name: Compile workflows
        env:
          GH_AW_REQUIRE_SELF_HOSTED_RUNNERS: "true"
        run: gh aw compile

When enabled, compilation checks the generated workflow jobs, including framework jobs and jobs whose runner comes from a default. Each job must include the self-hosted label or select a runner group. Custom labels alone are not sufficient because the compiler cannot verify that they identify self-hosted runners. An organization can make this check mandatory with a centrally managed required workflow or ruleset check.

gh aw env update defaults.yml --scope org --org MY_ORG --visibility all
gh aw env update defaults.yml --scope ent --enterprise MY_ENT --visibility all

Unified Session Logs

GH_AW_OTLP_LOGS_ENABLED=true opts in to content-minimized unified-session export from the conclusion job, before cleanup. Set it as an organization, enterprise, or repository Actions variable. Workflows need to be compiled with this support once; subsequent variable changes take effect at runtime. Unset or false disables export, and invalid values warn without exporting. Like other Actions variables, this is cooperative configuration, not a locked enterprise policy: repository variables and explicit workflow environment values can override it.

GH_AW_OTLP_LOGS_MAX_BYTES accepts a positive integer byte limit (default 65536, or 64 KiB). It limits the complete UTF-8 JSON request, including the OTLP envelope, for each vendor, not just the session body. Whole low-priority records are omitted until the request fits, and the exported included/omitted counts and truncation flag reflect those omissions. Invalid limits or limits too small for one record warn without exporting. Increasing the limit does not raise the compact profile’s independent source/event limits.

Both the compact serializer and OTLP envelope use this retention order:

Priority (highest first)	Records
1	Session format and workflow metadata
2	Final execution/session/job outcomes, plus failed, denied, exceeded, cancelled, or interrupted evidence from any retained event kind
3	Remaining guardrail decisions and grader results
4	Harness attempts/retries and engine reconnect diagnostics
5	Per-step usage and firewall token-usage observations
6	Successful memory updates

Later essential records can replace earlier low-priority records even after the compact byte/event limits are reached. Selection is deterministic: equal-priority records prefer earlier observations, and retained records remain in source order. Records are dropped whole, not text-truncated or merged; unrelated accounting sources are never summed. A GitHub Actions warning explicitly identifies a partial session export whenever output or source-scan limits discard evidence, and exported truncation/count attributes preserve that qualification for vendor queries. Routine privacy filtering alone does not trigger a size-truncation warning.

Logs use the existing OTLP endpoint list and per-endpoint authentication, including the enterprise default endpoint/header pair. No additional credentials or agent network access are required. The exporter posts OTLP/HTTP JSON or protobuf to <endpoint>/v1/logs, replacing a trailing /v1/traces or /v1/logs when present. OTEL_EXPORTER_OTLP_LOGS_PROTOCOL overrides OTEL_EXPORTER_OTLP_PROTOCOL for logs (http/json by default); OTEL_EXPORTER_OTLP_TRACES_PROTOCOL does not affect logs. The byte limit is measured against the complete JSON representation before encoding. For Azure Monitor traces endpoints, the logs URL also selects the independent Microsoft-OTLP-Logs stream, retaining the DCE, DCR, and query parameters. The vendor must support OTLP logs as well as traces at that base URL. See protobuf encoding and Azure Monitor.

Each retained event is a structured log body with type, data, optional timestamp, and provenance. Filter on the gh-aw.session.event_type log attribute or the gh-aw.session instrumentation scope. Records carry the existing trace/span IDs when valid, so vendors can correlate session metadata with traces. Resource attributes report the compact format version, included and omitted event counts, oversized source lines, and truncation.

timeUnixNano contains the original event time when valid; timezone-qualified ISO timestamps retain up to nanosecond precision. Numeric timestamps reuse the source-normalized provenance.timestampMs clock when present, otherwise the native millisecond timestamp. observedTimeUnixNano records export observation time independently. Missing, invalid, negative or out-of-range event times leave timeUnixNano unset; they are not replaced with export time. The body retains the projected original timestamp and available normalized clock.

Use the following resource attributes to qualify vendor queries:

Attribute	Meaning
gh-aw.session.compact.version	Compact profile version, currently 1.
gh-aw.session.included_events	Records retained in this request.
gh-aw.session.omitted_events	Inspected records omitted by privacy/vocabulary filtering or compact/transport budgets; not a count of the unscanned population.
gh-aw.session.oversized_lines	Completed source lines discarded for exceeding the line limit.
gh-aw.session.scan_truncated	Source scanning stopped at its byte limit.
gh-aw.session.truncated	Output, line-size or scan limits discarded evidence; false does not certify full-session completeness.
gh-aw.session.untimed_events	Retained records without an exportable event timestamp.

These attributes describe a request and repeat on its log rows. Do not sum them per row or across mirrored endpoints. A truncated or scan-limited export is partial evidence, not proof of no further activity. Event-time queries must qualify untimed records separately rather than silently treating export time as execution time. Uncorrelated records cannot establish run-level coverage.

The export reuses the compact unified-session profile: at most 128 events and 16 KiB of projected JSONL, with bounded source scanning. Prompts, conversation, reasoning, tool arguments/results, unknown event types, and unbounded nested data are excluded. Export additionally removes repository, workflow/run/session/agent identifiers, refs, commit SHAs, memory identifiers, branches, and filesystem paths. Current publication-secret and exporter-header masks are applied before sending. Operational metadata such as model names, job/grader IDs, timestamps, usage, outcomes, and trace correlation remains; this is content minimization, not a guarantee of irreversible anonymity. The full session artifact is unchanged. Export failures warn and do not change workflow results; authorization failures are not retried.

Artifact Retention

artifact_retention_days in .github/workflows/aw.json sets retention for all artifacts uploaded by compiled gh-aw workflows, including logs, inter-job staging, memory artifacts, safe-output uploads, custom actions/upload-artifact steps, and maintenance artifacts:

.github/workflows/aw.json

{
  "artifact_retention_days": 7
}

The field accepts an integer from 1 to 400 or a single-line GitHub Actions expression that resolves to an integer in that range, such as "${{ vars.ARTIFACT_RETENTION_DAYS || '7' }}". Expressions must use contexts available in every uploading job; prefer vars rather than job-specific steps or needs outputs. Recompile workflows after changing aw.json.

Retention precedence is:

1. artifact_retention_days in .github/workflows/aw.json
2. vars.GH_AW_DEFAULT_ARTIFACT_RETENTION_DAYS at runtime
3. The existing per-artifact setting or built-in default

Both repository and enterprise settings override individual retention-days values. Without either setting, existing retention is preserved, including short-lived staging artifacts, the safe-output upload default of 30 days, and GitHub’s repository default for uploads without an explicit retention period.

Set GH_AW_DEFAULT_ARTIFACT_RETENTION_DAYS as a repository, organization, or enterprise Actions variable, or manage it through gh aw env using default_artifact_retention_days: "7" in the defaults file. The variable is resolved at workflow runtime, so changing it does not require recompilation of workflows already compiled with this support.

GitHub’s repository, organization, and enterprise retention limits still apply (public repositories allow up to 90 days; private repositories allow up to 400). This setting does not change Actions cache eviction, git-backed memory, published assets, or artifacts uploaded inside third-party actions.

Project Timezone

By default, the CLI renders timestamps (table output, expiration footers, and the closing messages on expired issues, pull requests, and discussions) using the runner’s local clock. Set a project home UTC offset so these times render consistently regardless of where the CLI runs.

Configure the offset per repository with the utc field in .github/workflows/aw.json:

{
  "utc": "-08:00"
}

The value must be a numeric UTC offset in +HH:MM or -HH:MM form (for example +00:00, +05:30, or -08:00), within the range -14:00 to +14:00. Named timezones and abbreviations are not accepted.

To set an organization- or enterprise-wide default, use the GH_AW_DEFAULT_UTC environment variable (or the default_utc key managed by gh aw env). The repository aw.json value takes precedence over this enterprise default.

When neither is configured, timestamp formatting is left unchanged and uses the runner’s local time.

Precedence

For model selection, precedence is:

1. engine.model in workflow frontmatter
2. GH_AW_MODEL_AGENT_* or GH_AW_MODEL_DETECTION_*
3. GH_AW_DEFAULT_MODEL_*
4. Built-in compiler fallback

For max AI credits, precedence is:

1. max-ai-credits in workflow frontmatter (compile-time literal)
2. max-ai-credits from imported shared workflows (compile-time, first-wins across imports)
3. vars.GH_AW_DEFAULT_MAX_AI_CREDITS GitHub Actions variable (action runtime)
4. Built-in constant default: 1000 AIC

The compiler emits ${{ vars.GH_AW_DEFAULT_MAX_AI_CREDITS || '1000' }} in a runtime patch script when no frontmatter or imported value is set, so the organization variable is resolved at workflow run time by the GitHub Actions runner — not at compile time. A value of -1 disables AWF budget steering at runtime. Positive values accept K/M suffixes such as 100M.

For max turn cache misses, precedence is:

1. max-turn-cache-misses in workflow frontmatter
2. GH_AW_DEFAULT_MAX_TURN_CACHE_MISSES
3. Built-in constant default: 5

The compiler emits apiProxy.maxCacheMisses directly in the AWF config JSON. When max-turn-cache-misses is omitted, the compiler reads GH_AW_DEFAULT_MAX_TURN_CACHE_MISSES from its process environment and falls back to 5 if the variable is unset or invalid.

For threat-detection max AI credits, precedence is:

1. safe-outputs.threat-detection.max-ai-credits in workflow frontmatter (compile-time literal)
2. vars.GH_AW_DEFAULT_DETECTION_MAX_AI_CREDITS GitHub Actions variable (action runtime)
3. Built-in constant default: 400 AIC

The compiler emits ${{ vars.GH_AW_DEFAULT_DETECTION_MAX_AI_CREDITS || '400' }} for threat-detection runs when safe-outputs.threat-detection.max-ai-credits is unset, so the organization variable is resolved at workflow run time by the GitHub Actions runner — not at compile time. A value of -1 disables AWF budget steering for detection runs at runtime. Positive values accept K/M suffixes such as 100M.

For daily AI credits workflow guardrails, precedence is:

1. max-daily-ai-credits in workflow frontmatter (compile-time literal)
2. max-daily-ai-credits from imported shared workflows (compile-time, first-wins across imports)
3. vars.GH_AW_DEFAULT_MAX_DAILY_AI_CREDITS GitHub Actions variable (action runtime)
4. Built-in constant default: 5000 AIC

The compiler emits ${{ vars.GH_AW_DEFAULT_MAX_DAILY_AI_CREDITS || '5000' }} when no frontmatter or imported value is set, so the organization variable is resolved at workflow run time by the GitHub Actions runner — not at compile time. A value of -1 in frontmatter explicitly disables the guardrail. Positive values accept K/M suffixes such as 100M.

For the generated agent job timeout, precedence is:

1. jobs.agent.timeout-minutes in workflow frontmatter
2. vars.GH_AW_DEFAULT_AGENT_JOB_TIMEOUT_MINUTES
3. Built-in compiler default of 60 minutes

The generated detection job timeout follows the same chain with jobs.detection.timeout-minutes, vars.GH_AW_DEFAULT_DETECTION_JOB_TIMEOUT_MINUTES, and a built-in default of 10 minutes. The agentic_execution step timeout uses top-level timeout-minutes, vars.GH_AW_DEFAULT_TIMEOUT_MINUTES, and a built-in default of 20 minutes.

For OTLP observability, precedence is:

1. observability.otlp in workflow frontmatter
2. observability.otlp from imported shared workflows
3. secrets.GH_AW_DEFAULT_OTLP_ENDPOINT, falling back to vars.GH_AW_DEFAULT_OTLP_ENDPOINT, with secrets.GH_AW_DEFAULT_OTLP_HEADERS (action runtime)

The compiler always emits OTLP environment variables. When no endpoint is configured in frontmatter or an import, it emits ${{ secrets.GH_AW_DEFAULT_OTLP_ENDPOINT || vars.GH_AW_DEFAULT_OTLP_ENDPOINT }} and ${{ secrets.GH_AW_DEFAULT_OTLP_HEADERS }} so a repository or organization can enable telemetry for every agentic workflow without editing individual workflows. The endpoint secret takes precedence over the variable; clear both the endpoint secret and variable to disable export. When both are unset, the endpoint resolves to an empty string and OTLP export becomes a no-op; a configured endpoint without the matching headers secret is dropped by every span-emitting job (setup, conclusion, outcome, and MCP gateway) instead of being exported unauthenticated, and the agent job additionally fails the run so the misconfiguration is visible. If a workflow’s own env: block already defines one of the OTLP variables, the compiler skips injecting that variable rather than emitting a duplicate mapping key.

For detection engine selection, precedence is:

1. safe-outputs.threat-detection.engine in workflow frontmatter
2. Main workflow engine (engine)
3. Built-in compiler default

For detection model selection, precedence is:

1. safe-outputs.threat-detection.engine.model in workflow frontmatter
2. GH_AW_DEFAULT_DETECTION_MODEL
3. Engine-specific detection defaults

For project timezone (rendered CLI timestamps), precedence is:

1. utc in .github/workflows/aw.json
2. GH_AW_DEFAULT_UTC
3. The runner’s local clock (formatting left unchanged)

Example

Set an org-wide Codex model fallback:

gh variable set GH_AW_DEFAULT_MODEL_CODEX --org my-org --body "gpt-5.5"

Set an org-wide default max-ai-credits guardrail:

gh variable set GH_AW_DEFAULT_MAX_AI_CREDITS --org my-org --body "15M"

gh variable set GH_AW_DEFAULT_MAX_AI_CREDITS --org my-org --body "100M"

Set an org-wide default detection max-ai-credits guardrail:

gh variable set GH_AW_DEFAULT_DETECTION_MAX_AI_CREDITS --org my-org --body "750"

Set an org-wide default daily workflow AIC guardrail:

gh variable set GH_AW_DEFAULT_MAX_DAILY_AI_CREDITS --org my-org --body "15M"

Set an organization-wide GitHub Actions variable for the timeout and compiler process defaults for max-turns:

gh variable set GH_AW_DEFAULT_TIMEOUT_MINUTES --org my-org --body "30"
gh variable set GH_AW_DEFAULT_AGENT_JOB_TIMEOUT_MINUTES --org my-org --body "90"
gh variable set GH_AW_DEFAULT_DETECTION_JOB_TIMEOUT_MINUTES --org my-org --body "15"
export GH_AW_DEFAULT_MAX_TURNS=12
export GH_AW_DEFAULT_MAX_TURN_CACHE_MISSES=7
export GH_AW_DEFAULT_DETECTION_MODEL=gpt-5.5-mini

Set an org-wide default project timezone (Pacific Standard Time):

gh variable set GH_AW_DEFAULT_UTC --org my-org --body "-08:00"