Shared Telemetry Overlay

When to Apply

Activates whenever the parent agent produces or revises artifacts that touch observable behavior, audit trails, or production telemetry decisions.

Required Vocabulary Source

Always consult the telemetry-foundations skill for trace, metric, log, PII, and resource-attribute vocabulary. Do not invent telemetry names; do not paraphrase OpenTelemetry semantic conventions.

Decision Tree

1. Is the new behavior observable in production? If no, stop. No telemetry required.
2. Does it cross a service boundary or process? If yes, require trace span(s) per the skill's Trace Vocabulary section.
3. Does it produce a measurable rate, count, or duration? If yes, choose an instrument from the skill's Metric Vocabulary section and apply UCUM units.
4. Does it carry PII? If yes, consult the skill's pii-denylist reference and apply the redaction strategy listed there.
5. Is the cardinality bound? If no, demote to a log event; do not emit as a metric attribute.
6. Apply the artifact-specific mandatory telemetry from the table below that matches the artifact context.

Artifact-Specific Mandatory Telemetry

Artifact context	Additional mandatory telemetry
SSSC plans (sssc-plans/)	Require build/release telemetry attributes (vcs.*, ci.*) on supply-chain controls per the skill's Resource Attributes section.
SSSC review reports (sssc-reviews/)	Flag supply-chain findings that rely on build/release telemetry without corresponding vcs.* or ci.* evidence.
RAI plans (rai-plans/)	Capture model-output telemetry (latency, refusal rate, content-filter triggers) as metrics in the impact-assessment record.
Security plans (security-plans/)	Treat security-event emission as mandatory; cross-reference STRIDE entries with the skill's Log Vocabulary severity levels.
ADR artifacts (adr-plans/, docs/planning/adrs/)	Record the chosen telemetry strategy under "Consequences"; cite the skill section that justifies each instrument choice.
PRD sessions (prd-sessions/)	Capture telemetry acceptance criteria in the PRD's "Success Metrics" and "Operational Readiness" sections.
Accessibility plans (accessibility/)	No additional mandate beyond steps 1-5; apply the decision tree to any observable accessibility behavior.
Privacy plans (privacy-plans/)	Capture data-processing telemetry decisions, consent-state transitions, and retention/erasure events as auditable log or metric signals when they are observable in production.
Privacy review reports (privacy-reviews/)	Flag privacy findings that involve data-processing or consent-state signals without corresponding auditable log or metric evidence.
Code-review reports (reviews/code-reviews/)	Flag any production code path that emits telemetry without a corresponding semantic-convention reference.
Implementation changes (changes/)	Verify each new emitter's attributes against the skill before marking the implementation step complete.

Fallback

When the skill does not yet cover a needed concept, propose an addition through the skill's proposed-additions reference in the same change. Do not silently invent vocabulary.