visado: observability, UI, launch research (2026-11-18)
Observability: no OTel semconv for identity/authz exists yet. visado can reuse gen_ai.agent.*, gen_ai.tool.name and mcp.* (all Development, now in the separate semantic-conventions-genai repo), put the chain under a visado.* namespace, and track GenAI PR #291 (agent authz attributes). enduser.id was deprecated in core v1.45 and moved to the client-side repo.
Rust: opentelemetry 0.33.1 (2026-10-08). Logs and metrics are stable, traces are still beta. Pin opentelemetry* and pair it with tracing-opentelemetry 0.34.
Audit: OCSF API Activity (6003) with actor.authorizations and actor.invoked_by. CAEP 1.0/SSF 1.0 are Final (Sept 2025) and fit revocation signals. AIMS needs tamper-evident audit records. PR #176 keeps the transcription mapping for audit.
UI: a static SPA embedded in the binary, using React Flow with a left-to-right dagre/ELK layout, one hop per node, and allow/deny coloring. The target screenshot is one request with the denied hop lit up red.
Launch: agent-identity Show HNs score low (Zalando AIB 25 pts, Nov 2026). Projects that took off had a hook: a threat model (Agent Vault 156), a story (Cloudflare “Claude wrote it”), research first (mcp-scan), or a foundation donation at a conference (agentgateway). Proposed window: IETF 128 hackathon (Mar 6-7, 2027, inferred), then KubeCon EU, then AGNTCon NA.
1. Observability
Section titled “1. Observability”OTel semconv state
Section titled “OTel semconv state”- The GenAI and MCP attributes moved to
open-telemetry/semantic-conventions-genai, and the core registry marksmcp.*andgen_ai.agent.idas “Deprecated: moved”. Everything there is Development stability. https://opentelemetry.io/docs/specs/semconv/registry/attributes/mcp/ , https://github.com/open-telemetry/semantic-conventions-genai - Agent spans are
create_agent(CLIENT),invoke_agent/invoke_workflow(INTERNAL),plan, and execute_tool.gen_ai.agent.idmust be a provider-assigned stable ID, not an in-memory instance ID.gen_ai.conversation.idis set only when a real ID exists. The spec has no guidance on span links and defines no identity attributes. https://github.com/open-telemetry/semantic-conventions-genai/blob/main/docs/gen-ai/gen-ai-agent-spans.md - MCP spans are named
{mcp.method.name} {target}(for exampletools/call get_weather) and carrymcp.session.id,mcp.protocol.version,jsonrpc.request.id, andgen_ai.tool.name. Metrics aremcp.{client,server}.operation.duration. The spec recommends these over the RPC semconv. https://github.com/open-telemetry/semantic-conventions-genai/blob/main/docs/gen-ai/mcp.md enduser.*: v1.45.0 deprecatedenduser.id/enduser.pseudo.idin core and moved them to semantic-conventions-client-side (v0.1.0-dev).enduser.rolemaps touser.roles.enduser.scopewas removed with no replacement. https://github.com/open-telemetry/semantic-conventions/releases/tag/v1.45.0 , https://github.com/open-telemetry/semantic-conventions/pull/4238- Authz attributes are only proposals so far. GenAI issue #180 and PR #291 add
gen_ai.agent.*trust/authz attributes and deferfga.*decision-path attributes. Their invariant is that a deny is a span with no child execute span. OpenLLMetry PR #4701 proposesgen_ai.authz.decision. https://github.com/open-telemetry/semantic-conventions-genai/pull/291 , https://github.com/traceloop/openllmetry/pull/4701- Inference: visado could become the reference emitter and comment on #291 with a chain-aware shape.
How to emit a chain (my recommendation; not in any spec)
Section titled “How to emit a chain (my recommendation; not in any spec)”- Use one SERVER span per request,
visado.authorize {gen_ai.tool.name}, with these attributes:visado.chain.depth(int)visado.chain.subjects(string[]:human:alice,agent:planner,agent:coder,tool:github)visado.decision(allow|deny)visado.deny.hop(int)visado.txn_token.jtiauthzen.policy.id- plus
gen_ai.agent.name,gen_ai.tool.name,mcp.method.name,mcp.session.id
- Add one span event per hop (
visado.hop) carrying the subject, act claim, token type, scope narrowing and hop verdict. Events have timestamps, which plain attributes lack. https://grafana.com/blog/2024/08/15/all-about-span-events-what-they-are-and-how-to-query-them/ - Add a span link to the upstream agent’s span context when an inbound
traceparentexists or the token carries a trace ref, so cross-trace delegation stays navigable. - Tempo TraceQL can query span, event and link scopes, for example
{ event:name = "visado.hop" }and{ link:traceID != "" }. https://grafana.com/blog/grafana-tempo-2-6-release-performance-improvements-and-new-traceql-features/ , https://grafana.com/docs/tempo/latest/traceql/construct-traceql-queries/ - Jaeger and Tempo show events and links as a list, not a graph, which is why visado needs its own UI (section 2).
Rust SDK maturity
Section titled “Rust SDK maturity”- opentelemetry-rust 0.33.1 (Oct 8 2026): OTLP logs and metrics exporters are stable, traces are beta. The 1.0 planned for June 2025 has slipped. https://opentelemetry.io/docs/languages/rust/ , https://github.com/open-telemetry/opentelemetry-rust/blob/main/docs/migration_0.28.md
- tracing-opentelemetry (tokio-rs) is versioned separately: 0.34 pairs with otel 0.33. Expect breaking changes, so bump everything in lockstep. https://docs.rs/tracing-opentelemetry
- The claim that 0.33.1 made logs/metrics OTLP stable comes from search-result summaries. Check it against the GitHub release notes before quoting.
Audit formats
Section titled “Audit formats”- AIMS (draft-ietf-wimse-aims-01, 2026-11-02): audit, delegation and authz depend on a stable WIMSE identifier, and audit records MUST be tamper-evident. https://datatracker.ietf.org/doc/draft-ietf-wimse-aims/
- PR #176 (nik-kale) adds privacy text. Where stable IDs are not acceptable, the AS uses claims transcription (OAUTH-ID-CHAIN section 2.5) and keeps the mapping for audit. During review the text was softened to defer to deployment policy. Merge status is not visible. https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/pull/176
- Implication: visado’s audit record should store both pre- and post-transcription subject IDs, gated by policy.
- CAEP 1.0 / SSF 1.0 / RISC 1.0 are Final (vote Sept 2025). An interop profile is in progress. Use them as an output, for example emitting
session-revoked/token-claims-changeSETs when a chain is revoked. https://openid.net/three-shared-signals-final-specifications-approved/ - OCSF:
- API Activity (6003) is the best fit per request.
actor.authorizations[]carries decision and policy, andactor.invoked_byis a natural slot for the delegator. https://schema.ocsf.io/1.3.0/objects/actor - Authorize Session (3003) is session-level only.
- Open issue #1736 debates mapping agent behavior to OCSF, so there is no settled agent mapping yet. https://github.com/ocsf/ocsf-schema/issues/1736
- Inference: put the full chain in
unmappedor a profile until OCSF settles.
- API Activity (6003) is the best fit per request.
How comparable tools show access paths
Section titled “How comparable tools show access paths”- Teleport Graph Explorer:
- Paths always run left to right: Users, User Groups, Actions, Resource Groups, Resources.
- It shows allow and deny paths, and deny wins.
- Right-click filters the graph to one path.
- There is also a SQL editor and a
tsh access-graphCLI. - Enterprise only.
- https://goteleport.com/docs/identity-security/usage/graph-explorer/ , https://goteleport.com/docs/identity-security/
- Astrix: an “Identity Graph” of agents, MCP servers, NHIs, secrets and owners, with each agent mapped to a human owner. https://astrix.security/product/ai-agent-discovery/
- Aembit: no graph view found, only a “blended identity” audit trail. The SACR/buyer guidance says to ask for a two-hop demo where the final tool log retains the full chain. https://aembit.io/blog/human-vs-ai-identity-why-ai-agents-are-breaking-identity/ , https://softwareanalyst.substack.com/p/emerging-agentic-identity-access
- Inference: this exact demo is visado’s screenshot.
- OpenFGA Playground:
- A check is shown visually as the tuple path that proves it (Expand API).
- Community tools add a step-by-step “proof” view and a share-link that encodes the whole store in the URL (no backend).
- OpenFGA Studio uses React Flow.
- https://docs.fga.dev/intro/playground , https://github.com/hiren0ded/fga-playground
- Kiali dropped Cytoscape for PatternFly Topology (2.0 deprecated it, 2.8 removed it). https://kiali.io/news/release-notes/
- Hubble UI: a React SPA plus a Go backend, with live flows drawn as animated edges. The graph library is not confirmed. https://github.com/cilium/hubble-ui
What makes a screenshot compelling (synthesis)
Section titled “What makes a screenshot compelling (synthesis)”- One request, a linear left-to-right chain, and each edge labeled with the token step (ID token, then ID-JAG/OBO, then Txn-Token) plus the scope narrowing (
repo:*torepo:foo:read). - The deny hop highlighted red, with the AuthZEN reason inline.
- A side panel with the decoded Txn-Token and the AuthZEN request/response.
- A “copy as TraceQL / open in Grafana” link.
Recommendation
Section titled “Recommendation”- Serve a static SPA from the gateway, embedded with
rust-embed/include_dirat/ui. It reads a JSON/api/requests/{id}/chainendpoint and an SSE live tail. - Use React Flow (
@xyflow/react) with a dagre/ELK LR layout. Chains are small DAGs, so Cytoscape’s large-graph strengths are not needed. - Add a share-by-URL mode (chain JSON in the fragment, as fga-playground does) so people can paste chains into issues.
- Add a
visado explain <req-id>TUI/CLI twin, similar to Teleport’stsh access-graphand openfga-tui.
3. Launch playbook
Section titled “3. Launch playbook”Evidence (HN, via Algolia)
Section titled “Evidence (HN, via Algolia)”- Infisical “Show HN: Agent Vault” (2026-04-22): 156 pts. https://news.ycombinator.com/item?id=47865822
- The post had a clear threat model (credential vs data exfiltration), cited prior art (Anthropic/Vercel/Cloudflare brokering), listed design decisions as bullets, admitted the DX was rough, and asked for feedback.
- It had no live demo and leaned on Infisical’s brand.
- Cloudflare workers-oauth-provider: not a Show HN. The thread “Cloudlflare builds OAuth with Claude and publishes all the prompts” (June 2025) went viral on the AI-authorship story, and a critique post followed. https://news.ycombinator.com/item?id=44159166 , https://news.ycombinator.com/item?id=44215667
- I could not verify the 889 pts figure.
- Better Auth:
- “Launch HN” (YC P25, May 2025): 259 pts.
- Third-party posts did better: Val Town’s migration post got 304 pts, and the TechCrunch raise got 282.
- Earlier self-posts got 10 to 73 pts.
- https://hn.algolia.com/api/v1/search?query=better%20auth&tags=story
- mcp-scan: launched 2025-04-11 right after Invariant’s tool-poisoning disclosures. Its HN posts got 1 to 6 pts, yet it reached about 3.2k stars and Snyk acquired the company (now snyk/agent-scan). Cisco’s mcp-scanner got 168 pts. https://invariantlabs.ai/blog/introducing-mcp-scan , https://github.com/invariantlabs-ai/mcp-scan
- agentgateway: Rust, contributed to the LF at Open Source Summit EU on 2025-08-25, and now under AAIF. Its traction came from foundation and enterprise quotes, not HN (max 2 pts). https://www.opensourceforu.com/2025/09/linux-foundation-strengthens-open-source-agentic-ai-with-solo-ios-agentgateway/
- ToolHive: negligible HN presence (2 pts).
- Same category as visado:
- Zalando “Agentic Identity Broker” Show HN (2026-11-06): 25 pts. It integrates with agentgateway via RFC 8693, uses a one-command
just compose-up-detached, and asked for UI/UX feedback. https://github.com/zalando-incubator/agentic-identity-broker - Agent Passport: 14 pts.
- Zalando “Agentic Identity Broker” Show HN (2026-11-06): 25 pts. It integrates with agentgateway via RFC 8693, uses a one-command
Takeaways
Section titled “Takeaways”- Inference: a bare “agent identity gateway” Show HN will probably land at 10 to 30 pts.
- A hook works better:
- (a) a research post, such as “we replayed N agent delegation chains and found scope widening”, mcp-scan style; or
- (b) a visual: an HN-native title showing the chain, for example “Show HN: Visado – see every hop of an AI agent’s delegation chain (Rust)”.
- Title format: “Show HN: Name – plain-language what it does”, with no buzzword stack.
- README order:
- A GIF/asciinema of the chain UI above the fold
docker compose up/cargo installwith a one-command demo that runs human, agent, agent and tool locally, and ends on a deny- “Why” (threat model, 3 bullets)
- Specs implemented (Txn-Token, AuthZEN, WIMSE, AIMS) with versions
- Compare table (agentgateway, AIB, Aembit)
- Status/limits
- Interop with agentgateway and AIB adds credibility, since AIB already pairs with agentgateway.
Calendar
Section titled “Calendar”- IETF 127 SF is Nov 14-20, 2026 (now). https://www.ietf.org/meeting/upcoming/
- IETF 128 Kuala Lumpur is Mar 6-12, 2027. The hackathon is probably Mar 6-7; that is inferred from the usual Sat-Sun pattern and not confirmed. http://www.mail-archive.com/ietf-announce@ietf.org/msg26949.html
- IETF 126’s hackathon drew about 800 registrants and about 70 projects. https://www.ietf.org/blog/ietf126-recap/
- KubeCon EU 2027 is in Barcelona, Mar 15-18. https://contribute.cncf.io/events/cfp/
- The main CFP closed Oct 11, 2026, and notifications go out Dec 7.
- The co-located CFP reportedly closed Oct 18.
- Maintainer Summit is Mar 14. Rejekts is Mar 13-14.
- “Agentics Day: MCP + Agents Europe” is Mar 15 in Barcelona.
- AGNTCon + MCPCon NA (formerly MCP Dev Summit) is Apr 28-29, 2027 at Moscone SF. Europe is Sep 15-16, 2027 in London. https://events.linuxfoundation.org/agntcon-mcpcon-north-america-2027/
- Suggested sequence:
- Txn-Token/AIMS interop at the IETF 128 hackathon (results slide plus WG list post)
- Show HN on a Tue-Wed morning US time, about a week later and away from conference noise (inference)
- Booth or lightning demos at Rejekts/Agentics Day
- Submit to the AGNTCon CFP (check its deadline)
CNCF sandbox day-1 hygiene (CLOMonitor checks)
Section titled “CNCF sandbox day-1 hygiene (CLOMonitor checks)”- An
ADOPTERS*file - DCO: enable the DCO app on all repos
- OpenSSF Best Practices badge in README
- Scorecard badge in README (
api.scorecard.devURL) - Scorecard-derived checks: binary artifacts, code review, dangerous workflow, dependency update tool, maintained, signed releases, token permissions
- A recent release
- An approved license (Apache-2.0 is OK) and license scanning
- A CNCF Slack reference in README
- An OpenSSF Security Insights manifest
- SECURITY.md, CODE_OF_CONDUCT, GOVERNANCE, MAINTAINERS, CONTRIBUTING
- Sources: https://clomonitor.io/docs/topics/checks/ , https://github.com/cncf/clomonitor/blob/main/docs/checks.md
- The sandbox onboarding asks for the Best Practices badge, while the security guidelines push Scorecard, so both are needed.
- For a Rust project, also:
cargo-deny/cargo-auditin CI, signed releases (cosign/sigstore), SLSA provenance, and pinned action SHAs (helps the Scorecard token-permissions and pinned-deps checks).
Gaps / confidence
Section titled “Gaps / confidence”- The 889 pts figure for workers-oauth-provider is not verified.
- Hubble UI’s graph library is unknown.
- The IETF 128 hackathon exact days are inferred.
- AGNTCon NA CFP deadline was not checked.
- The otel-rust 0.33.1 stability claim needs a check against the release notes.
- Everything in the GenAI/MCP semconv is Development and still churning.
© 2026 The visado Authors. Licensed under Apache-2.0.
Third-party logos are trademarks of their respective owners and are used to show compatibility; their use does not imply endorsement.