Add AI credit cap observability attributes to OTLP conclusion spans

[Copilot]

This change extends OTLP conclusion-span telemetry to include AI credit cap context, so runs can be distinguished by budget configuration and cap-related failure signals alongside existing gh-aw.aic cost data. It surfaces cap value and cap/rate-limit state already resolved in ai_credits_context.cjs without adding provenance or unrelated billing metadata.

• Conclusion-span AI credit cap attributes

  • send_otlp_span.cjs now imports and uses resolveAICreditsFailureState().
  • Emits gh-aw.max_ai_credits as a numeric attribute only when a valid value is available.
  • Emits gh-aw.max_ai_credits_exceeded and gh-aw.ai_credits_rate_limit_error as boolean attributes on conclusion spans.

• Scope and behavior constraints preserved

  • Existing gh-aw.aic behavior is unchanged.
  • No source/provenance fields (e.g., gh-aw.ai_credits.source) were added.
  • New attributes are attached to conclusion spans, not unrelated spans (including dedicated agent spans).

• Test coverage updates

  • Added assertions in send_otlp_span.test.cjs for:
    • numeric emission of gh-aw.max_ai_credits
    • boolean emission of gh-aw.max_ai_credits_exceeded
    • boolean emission of gh-aw.ai_credits_rate_limit_error
    • omission of malformed/missing gh-aw.max_ai_credits
    • conclusion-only placement of new fields

• Observability docs/spec alignment

  • Updated docs/src/content/docs/reference/open-telemetry.mdx attribute table.
  • Updated specs/otel-observability-spec.md conclusion-span conditional attributes.

const { maxAICredits, aiCreditsRateLimitError, maxAICreditsExceeded } = resolveAICreditsFailureState();

const maxAICreditsValue = normalizeNonNegativeNumber(maxAICredits);
if (typeof maxAICreditsValue === "number") {
  attributes.push(buildAttr("gh-aw.max_ai_credits", maxAICreditsValue));
}
if (typeof maxAICreditsExceeded === "boolean") {
  attributes.push(buildAttr("gh-aw.max_ai_credits_exceeded", maxAICreditsExceeded));
}
if (typeof aiCreditsRateLimitError === "boolean") {
  attributes.push(buildAttr("gh-aw.ai_credits_rate_limit_error", aiCreditsRateLimitError));
}

[Copilot]

Pull request overview

This pull request extends the OTLP conclusion span telemetry emitted by actions/setup to include AI credit cap context (configured cap and cap/rate-limit failure signals), enabling downstream observability to distinguish runs by budget configuration alongside existing gh-aw.aic cost data.

Changes:

• Emit gh-aw.max_ai_credits, gh-aw.max_ai_credits_exceeded, and gh-aw.ai_credits_rate_limit_error as conclusion-span attributes via resolveAICreditsFailureState().
• Add Vitest assertions ensuring the new fields are emitted on conclusion spans (and not on dedicated agent spans), including omission behavior for invalid cap values.
• Update the OpenTelemetry reference docs and the observability spec to include the new attributes.

Show a summary per file

File	Description
actions/setup/js/send_otlp_span.cjs	Adds conclusion-span attribute emission for max AI credits and cap/rate-limit failure state.
actions/setup/js/send_otlp_span.test.cjs	Adds tests validating emission/omission and conclusion-only placement of the new attributes.
docs/src/content/docs/reference/open-telemetry.mdx	Documents the new OTel attributes in the reference table.
specs/otel-observability-spec.md	Updates the conclusion-span conditional attributes table to include the new fields.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 4/4 changed files
• Comments generated: 2

[mnkiefer]

@copilot Address all remaining review comments.

[Copilot]

@copilot Address all remaining review comments.

Done in commits b2a7b29 and 1fc3c4c:

Console logging noise (send_otlp_span.cjs:2363): Added a { logProvenance = true } options parameter to resolveAICreditsFailureState() in ai_credits_context.cjs. All [ai-credits] ... console.log lines are now guarded by if (logProvenance), so existing callers that rely on provenance logs are unaffected. The call in send_otlp_span.cjs passes { logProvenance: false } to keep OTLP exports silent.

Flaky test (send_otlp_span.test.cjs:3612-3614): Replaced the real-file approach (which was broken because mkdirSync is globally mocked in the describe block's beforeEach) with spies on the low-level fs calls consumed by parseAICreditsExceededFromAgentStdio — existsSync, statSync, openSync, readSync, and closeSync — each restored in a finally block.

[mnkiefer]

@copilot Fix tests

[Copilot]

@copilot Fix tests

Fixed in the latest commit. The root cause was a cross-test contamination in shard 1/4:

1. sendJobConclusionSpan (called from my new send_otlp_span.test.cjs tests) writes spans to /tmp/gh-aw/otel.jsonl, including "gh-aw.ai_credits_rate_limit_error" as an attribute key.
2. handle_agent_failure.cjs's hasEngineRateLimit429InOTELMirror() reads that file and applies ENGINE_RATE_LIMIT_429_RE. The pattern rate_limit_(?:error|exceeded) matched the substring rate_limit_error in the attribute key name — a false positive.
3. This caused buildEngineFailureContext() to detect a 429 and try to render a missing template, catching the exception and returning "".

Two fixes:

• Regex: Added \b word boundaries around rate_limit_(?:error|exceeded) so it no longer matches when rate_limit is part of a larger identifier like ai_credits_rate_limit_error.
• Test isolation: Added GH_AW_OTEL_JSONL_PATH to the buildEngineFailureContext with terminal_reason guard describe block's beforeEach/afterEach, following the same pattern already used in the sibling buildEngineFailureContext describe block.