docs: sync ADP changes from cloudv2 (2026-07-04)

[micheleRP]

Summary

Syncs one user-facing Agentic Data Plane change from cloudv2 into the docs: the new agent session permissions.

📄 Deploy preview: https://deploy-preview-125--redpanda-agentic-data-plane.netlify.app/agentic-data-plane/control/permissions-reference/

cloudv2 added a public SessionService (agents/{agent}/sessions/{session}, proto v1alpha1) for listing, viewing, and deleting a managed agent's persisted conversation sessions. It introduces three new permissions that the docs' permissions reference must enumerate:

Permission	Operation	Writer	Reader
dataplane_adp_agent_session_list	List an agent's conversation sessions (summaries only)	✓	✓
dataplane_adp_agent_session_get	View one conversation session, including its full message transcript	✓	✓
dataplane_adp_agent_session_delete	Delete an agent's conversation sessions	✓

Role grants are taken from writerRolePermissions / readRolePermissions in permissions_constants.go: list and get are granted to both Writer and Reader; delete is Writer-only.

What changed in modules/control/pages/permissions-reference.adoc

• New Agent session permissions section (between agent triggers and transcripts), with the three permissions, their operations, and role grants.
• A note that dataplane_adp_agent_session_get returns full conversation content (prompts, tool inputs/outputs, model output) to the general-purpose Reader role — unlike the transcript permissions, which require the dedicated TranscriptReader role. Grounded in the source comment in permissions_constants.go flagging that session_get sits at transcript-access trust level.
• Updated the source-provenance comment at the top of the page.

Source commits

• 7093932 — adp: move SessionService to adp-api as an AIP session resource (PR #27659, author @birdayz). Defines session.proto, the new dataplane_adp_agent_session_{list,get,delete} permissions, cedar schema, and role grants.

Reviewer

Added @birdayz (author of the source commit) as an optional reviewer for a source-accuracy check. Their approval is not blocking.

Changes considered but not documented

The rest of the last 24 hours of cloudv2 main was out of scope for docs:

• adp-ui frontend work (mobile-responsive layouts, home-page attention dismissals, session-picker UI, visual-baseline regens, table-molecule refactors) — UI-only, not documented per the sync policy.
• Agent-level OTel traces (52b3c15, 9716715, 4247889) — gated behind a default-off, internal cluster template knob (template.observability.traces_enabled); no customer-facing surface yet.
• SystemPolicyService list filter + page-size bounds (AIP-158/160) — API-list refinement with no existing doc surface in adp-docs.
• rpk ai CLI hint/auth/env refactors — internal wording and bug fixes; the rpk ai reference is auto-generated and not hand-edited here. cloudv2 characterized v0.2.18 as a maintenance release.
• Infra/security hardening — LaunchDarkly SDK v9 bump, sentry-go bump, log-redaction fixes, image-override map, devenv/CI changes.

🤖 Generated with Claude Code

https://claude.ai/code/session_01D6G3MiyR6SvPh9JfzMXECz

[netlify]

✅ Deploy Preview for redpanda-agentic-data-plane ready!

Name	Link
🔨 Latest commit	92ef378
🔍 Latest deploy log	https://app.netlify.com/projects/redpanda-agentic-data-plane/deploys/6a48f93590834000082fdab4
😎 Deploy Preview	https://deploy-preview-125--redpanda-agentic-data-plane.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[micheleRP]

[adp-docs PR critic]

Verdict: looks accurate. I reviewed the diff to modules/control/pages/permissions-reference.adoc adversarially against cloudv2 source and found no problems.

Source-accuracy checks (all confirmed against cloudv2 origin/main, commit b5d24247, which includes source commit 7093932 / PR #27659):

• Permission names & operations — pkg/permissions/permissions_constants.go defines all three verbatim: dataplane_adp_agent_session_list ("List an agent's conversation sessions (summaries only)"), dataplane_adp_agent_session_get ("View an agent conversation session including its full message transcript"), dataplane_adp_agent_session_delete ("Delete an agent's conversation sessions"). The table's operation wording matches. ✓
• Role grants — verified in the same file: writerRolePermissions contains all three (list, get, delete); readRolePermissions contains only list and get, not delete. So the table (Writer ✓✓✓ / Reader ✓✓ / delete Writer-only) is correct. ✓
• The NOTE (Reader can read full conversation content via session_get) — grounded. The source comment above the constants says session_get "returns full conversation payloads (prompts, tool inputs/outputs), so it belongs at transcript-access trust level," yet it is in readRolePermissions. The comparative claim ("unlike the transcript permissions") also holds: dataplane_adp_transcript_{get,list} live in the separate transcriptReaderPermissions var (TranscriptReader role), with a source comment stating transcript read access is "granted through a dedicated role rather than bundled into the broad Writer/Reader defaults." ✓
• Prose claims — verified against proto/public/cloud/redpanda/api/adp/v1alpha1/session.proto: "managed agents only; self-managed keeps no session store" (proto: "Managed agents only. Self-managed / BYOA agents have no session store; the RPCs reject them with FAILED_PRECONDITION"); "runtime creates sessions, so there is no create or update permission" (proto: "The agent runtime creates and writes sessions (there is no Create/Update here)"); resource pattern agents/{agent}/sessions/{session} and proto version v1alpha1. ✓

Standards / correctness: AsciiDoc is well-formed — [cols="2,2,1,1"] matches the 4-column header, the empty Reader cell on the delete row is valid, and the [NOTE]/==== block is correct. Section is placed between the agent-trigger and [[transcript-permissions]] sections as described. No broken xrefs or resource IDs introduced. Source-provenance comment updated appropriately.

Completeness: The PR documents exactly the user-facing surface introduced by the source commit (the three new permissions + the Reader-trust caveat). No gaps found.

No changes recommended.

---

Generated by Claude Code