Add docs for agentic attributes in Signals

[jborlase-snowplow]

This PR was generated automatically by Claude.

What changed?

Adds documentation for agentic attributes, a private preview Signals feature where an LLM decides a value for a user's session and returns its reasoning.

• New page docs/signals/agentic-attributes/index.md ("Define agentic attributes"): how evaluation works, creating an agentic attribute through the Signals API (fields, output types, triggers, evaluation policy), and publishing, updating, unpublishing, and deleting

• New page docs/signals/applications/agentic-attributes/index.md ("Agentic attributes", under "Use in your application"): reading a session's value and the response format

• Adds the retrieval page to the list in docs/signals/applications/index.md

• New section "Analyze agentic attribute values in your warehouse" on the definition page: each written value is tracked as an agentic_attribute self-describing event, rendered with SchemaProperties

Both new pages carry a private preview note.

Why?

Agentic attributes are built in the Signals registry, streaming engine, and AI engine, but aren't documented. All limits, defaults, endpoints, and response fields come from the signals-monorepo source.

Reviewer guidance

• Setup uses only the REST API for now. A Console UI exists but isn't covered yet. Adding Console and SDK tabs later would match the agentic contexts page

• The read endpoint (GET /api/v1/agentic_attribute) is served by the Go server. The page assumes it's on the same {{API_URL}} as the management API. Please confirm

• The published Python SDK (0.4.9) still uses the old decision_goal field name and has no publish or read helpers, so the pages don't document the SDK

• The private preview note says "Contact Snowplow to enable them". Swap in a specific channel if there is one

• Merge dependencies: the event section documents behavior from snowplow-incubator/signals-monorepo#764 and the schema from Add com.snowplowanalytics.signals/agentic_attribute/1-0-0 iglu-central#1488, neither merged yet. Don't publish before both ship. The schema prop is copied verbatim from Move autogen data models page to data models section #1488

AI reviews

Claude will automatically review this PR against the docs style guide.

If you have questions or want it to look again at something specific, tag @claude in a comment.

🤖 Generated with Claude Code

[claude]

Style review

These two pages are well written and closely follow the conventions. All internal links and anchors resolve. Frontmatter is complete. The diff adds no external URLs. I found a few small issues.

1. Release note: missing or deliberately skipped?

CLAUDE.md says: "New functionality needs a release note. If a docs change documents a new feature … add a release note in the same PR (or a follow-up PR, linked)." This PR documents a new feature, but it doesn't add a file under release-notes/. The feature is in private preview, so leaving out an announcement might be intentional. CLAUDE.md says to ask when unsure, so please confirm. If a note is needed, add release-notes/<slug>/index.md here or link a follow-up PR.

2. Dangling modifier in the intro (docs/signals/agentic-attributes/index.md)

Once defined and published, retrieve an agentic attribute in your application.

"Once defined and published" has no subject, so on a first read it attaches to the reader. This is a clarity issue. Suggested fix:

Once you've defined and published an agentic attribute, retrieve it in your application.

3. Passive voice where Signals is the actor

The style guide says: "Use active voice: 'the Collector receives events' not 'events are received by the Collector'." The pages mostly follow this, but a few sentences use the passive voice even though the actor is known. Example from docs/signals/applications/agentic-attributes/index.md:

Values are kept for seven days after they're written.

Suggested: "Signals keeps each value for seven days after writing it."

The same applies to "The model's answer, justification, and evaluation time are stored against the session's domain_sessionid" (step 4 of the evaluation list) → "Signals stores the model's answer, justification, and evaluation time against the session's domain_sessionid". Leave sentences like "a trigger still has to fire" unchanged.

4. Heading voice (minor)

CLAUDE.md says to "prefer imperative voice" in headings. ## How agentic attributes are evaluated is descriptive and passive. A possible alternative is ## Understand how Signals evaluates agentic attributes. You could also keep the current heading, since this is a conceptual section. I didn't flag ## Response format, because it matches the sibling agentic contexts page.

Everything else looks good: tables, list formatting, number style, admonition usage, and the update to the applications index.

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	documentation	7f56269	Commit Preview URL Branch Preview URL	Oct 09 2026, 03:33 PM