Two acceptance tests the spec names and nothing wrote: T483 and T486

[arpanghoshal]

These were pushed to release/0.10.0 a few minutes after #188 merged, so they missed it. Same
commit, on its own branch. Pairs with ctrlrun-docs readiness/6068, which moves the test count
this adds.

The audit that found them

Of T470 to T507, eight numbers had no test at all. Run, not remembered:

• Four are covered under another name and are a naming debt, not a coverage gap: T481 inside
  T474's not_mine case, T482 in test_hop_surfaces, T484 by the upgrade suite, T500 by
  T499a. T488 and T499 exist only as T488a/b/c and T499a.
• Two were genuinely uncovered, and are here.
• Two remain, named rather than left: T480 and T502. See the end.

T483 — a hop refusal's event carries §3.3's exact key set

As a set, and on both rows of the table. The negative half is the point: a refusal may carry
the id the caller already holds and the name of the row that stopped it, never the patterns,
the limits, the subject or the expiry. Otherwise each refusal is one question against the envelope
and a peer enumerates it a key at a time, which is what v0.9 §5.5 refuses when it lets a scope
hash reach a receipt while the scope never does.

== rather than not in, because assert "actions" not in data names one leak and passes over
every other. And both rows are read off §3.3's table, not off _authority_data — §11 records an
earlier version of this test asserting a key set that matched neither the prose nor the code.

Mutated: one permitted_actions key added to a hop refusal's payload. T483 fails. Restored, 28
pass.

T486 — a resume inside an unrelated ambient task= and hop= reads neither

v0.9 §6.3.2's ambient-context hazard, at the hop. A relay serving several agents resumes a
suspended leg from inside its own Control.hop(...) block; a resume that read the ambient value
decides agent A's suspended action against agent B's envelope. Fail-open whenever the ambient hop
is wider than the recorded one.

The first version of this test was not enough, and a mutation run is what said so. With the
event carrying both values as strings, a reader that falls back to the ambient value only when the
event's is absent takes the event's value anyway and passes. The added case is a 0.10 leg that
recorded {"task": null, "hop": null}: the keys are present, so it is not 0.9.0's silence, and the
answer must still be neither.

Mutated, twice. Ambient-as-fallback: fails. Ambient-takes-precedence: fails. Restored: 28 pass.
Only the second mutation failed against the first version of the test, which is why it was
rewritten.

What is still owed

• T480 — a hop presented with params.metadata.agent_id naming a different agent is decided
  on the IdentityProvider's principal. True by construction today: gateway/mcp.py reads only
  hop and task out of that bag and never agent_id. What is missing is the regression guard
  against a future read, not the property.
• T502 — the four v0.9 regressions as regression tests. §13.8 singles out its last row as the
  one nothing else in this repository would catch.

Full gate: 3,999 passed, 304 skipped. The one local failure is a half-installed langgraph in
the dev environment, not reproducible in CI's adapters job.

Not merging.

[coderabbitai]

Warning

Review limit reached

Next included review available in 5 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: c6a2f3ad-e8a8-4cd5-9e07-534f8198e7c0

📥 Commits

Reviewing files that changed from the base of the PR and between f3f9096 and 561b2cb.

📒 Files selected for processing (1)

• tests/test_hop.py

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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