feat(supervisor): hand back what a tick observed as a TickReport

[YellowSnnowmann]

Summary

Supervisor::tick now returns a TickReport — one SupervisorEvent per thing the cycle observed or did, in order — instead of (). The supervisor already logged all of it, but a log line is not something a host can route: it cannot be filtered into an event log a user reads, or turned into a notification when a server stays down. The one host that consumes this crate drives tick itself and had no way to learn that a probe timed out, that a session was torn down, or that a reconnect had been failing for an hour. That is what tinyhumansai/openhuman#5931 asks it to surface, and this is the crate-side half of it.

The supervisor still publishes no health signal of its own (the module docs' reasoning stands); it hands the observations back and the host decides which of them a user should hear about.

Related issue

tinyhumansai/openhuman#5931 — cross-repo, so no closing keyword. The host-side PR bumps vendor/tinymcp onto this commit and translates the report into that application's domain events and notifications.

API or behavior changes

Additive. Three new public types, re-exported from the crate root: TickReport { events }, SupervisorEvent (#[non_exhaustive]: ProbeAnswered, ProbeTimedOut, TransportDropped, Reconnected, ReconnectFailed, Parked), and ServerRef { server_id, qualified_name, display_name }. Supervisor::tick returns TickReport instead of (); every in-tree caller is in statement position, and there is deliberately no #[must_use] on the report so those stay valid. Supervisor::run drops the report.

Behaviour is unchanged: no log line moves or changes level, the streak rule and backoff are untouched. Reconnected::after_failures is read before the backoff entry is forgotten — that is the one new thing the cycle has to remember, and it is what lets a host tell a session rebuilt within the same cycle (nobody noticed) from a server that had stayed down.

Internal: the reconnect arm of tick moved into attempt_connect, and the timeout half of judge_probe into judge_timeout — the same split judge_probe itself got in #5, for the same reason, and to stay under clippy::too_many_lines.

Validation

Commands actually run, with their outcome:

• cargo fmt --all -- --check — passes.
• cargo clippy --all-targets --all-features -- -D warnings — fails on one pre-existing error that is not in this diff and fails identically on unmodified main (verified by stashing the change and re-running): unknown lint: clippy::unused_async_trait_impl at crates/tinymcp/src/tinybus_module/service.rs:122. This is the same toolchain artefact fix(supervisor): report what a probe observed instead of asserting a drop #5 recorded (local clippy 1.96 no longer knows that lint name). With only that lint name allowed (-D warnings -A unknown_lints) clippy reports nothing for this diff.
• cargo build --all-targets --all-features — passes.
• cargo test --all-features — passes: 678 + 4 + 149 + 12 + 18, 0 failed.
• .github/scripts/check-file-coverage.sh 90 coverage.json — passes; supervisor/report.rs 100%, supervisor/types.rs 91.86%, supervisor/test.rs 99.2%.

Tests

Nine new tests in registry/supervisor/test.rs, in a new "The report" section, over the existing adjustable fixture plus one new dial (fail_next_list: fail exactly one tools/list, so a probe finds a broken transport and the reconnect that follows can still complete — the shape of the field case in #5931, a transport that hiccups once and is fine a second later). Each branch of the cycle pins what it reports: an answered probe; a kept timeout with its place in the streak; the teardown after a run of timeouts and the refused reconnect behind it; a broken transport and the same-cycle rebuild with after_failures: 0; a failed reconnect with its penalty, and silence inside the backoff window; a recovery with after_failures: 1; a parked server reported once and then quiet; a disabled install and an empty store reporting nothing; ServerRef::from, kind() and server() on every variant.

Deliberately untested: the ProbeOutcome::Missing arm still (unchanged reasoning from #5 — forcing the entry to vanish between the membership check and the probe would test the harness, not the code); its report entry is covered by the every_event_names_its_server_and_its_kind construction instead.

Documentation

registry/supervisor/mod.rs gains a "What it hands back instead" section next to the existing "What it deliberately does not do", and report.rs documents every variant and field. No README or docs/ page describes the supervisor's API today, so none needed updating.

Checklist

• The change is focused on one logical change
• No new #[allow(...)], #[ignore], or relaxed lints
• No secrets, tokens, or .env contents in the diff or the description

🤖 Generated with Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Team

Run ID: 6c727c2b-90e0-4043-830e-f22e0d8a8011

Warning

Your free Security trial is over. An organization admin can upgrade to Advanced for continuous pull request security review or dismiss this notice.

---

_{Comment @coderabbitai help to get the list of available commands.}

[tinysweeper]

tinysweeper found nothing blocking. Approving.

             $0.0224 · 198,337 in / 2,321 out · 16,618 cached (8%) · deepseek/deepseek-v4-flash, openrouter/openai/text-embedding-3-small, z-ai/glm-5.2 · 737 embedded
critique:    $0.0094 · 77,832 in  / 855 out   · 8,090 cached (10%) · deepseek/deepseek-v4-flash, z-ai/glm-5.2
security:    $0.0090 · 77,706 in  / 881 out   · 8,528 cached (11%) · deepseek/deepseek-v4-flash, z-ai/glm-5.2
tests:       $0.0020 · 22,853 in  / 117 out   · 0 cached (0%)      · deepseek/deepseek-v4-flash
description: $0.0014 · 15,448 in  / 82 out    · 0 cached (0%)      · deepseek/deepseek-v4-flash

[tinysweeper]

How this change flows

4 changed behaviours across 16 relationships. 6 surrounding behaviours are shown (60 graph nodes walked). 34 further behaviours left out to keep the diagram readable.

flowchart LR
  n0["ServerDials<br/>changed"]:::changed
  n1["...nswers_with_an_error_is_torn_down_at_once<br/>changed"]:::changed
  n2["serve_adjustable_server<br/>changed"]:::changed
  n3["Supervisor<br/>changed"]:::changed
  n4["new"]:::impacted
  n5["tick"]:::impacted
  n6["connected_to"]:::impacted
  n7["install"]:::impacted
  n8["supervisor"]:::impacted
  n9["insert_server"]:::impacted
  n1 -->|calls| n2
  n1 -->|tests| n2
  n1 -->|calls| n4
  n1 -->|tests| n4
  n1 -->|calls| n5
  n1 -->|tests| n5
  n1 -->|calls| n6
  n1 -->|tests| n6
  n2 -->|uses| n0
  n2 -->|calls| n4
  n6 -->|calls| n4
  n6 -->|calls| n7
  n6 -->|calls| n9
  n7 -->|calls| n4
  n8 -->|uses| n3
  n8 -->|calls| n4
  classDef changed fill:#0d4429,stroke:#238636,color:#e6edf3
  classDef impacted fill:#161b22,stroke:#6e7681,color:#c9d1d9
  classDef flagged fill:#5a1e02,stroke:#d93f0b,color:#ffffff
  classDef blocking fill:#67060c,stroke:#f85149,color:#ffffff

Loading

_{Green: changed behaviour. Grey: surrounding behaviour. Arrows name the call, use, implementation, or test relationship. Orange: has findings. Red: has a finding that blocks the merge.}