[rigel-mintaka]

What does this PR do?

After #43, a smart-reviewer DENY goes to a human. Review found ways it could still run without a human answering it. Now a DENY runs only after a person answers "once" to that request; otherwise it is blocked.

Related Issue

RIG-8544. Follows #43.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• A DENY is never grouped with another request. Subagent auto-approve and bulk approve skip it.
• A session or always reply to a DENY counts as once.
• A reply must name a DENY by its ID to release it. A reply without an ID resolves nothing while a DENY is first in the queue, so cards that send no ID can't release or skip past it. /approve says to answer DENY requests on their own card.
• With review_first, the reviewer runs first; a cached grant then applies only to a non-DENY verdict.
• The CLI gets the DENY flag and always shows the warning and a command row, down to 11 terminal rows.
• A human deny or timeout adds the denial-breaker text; "once" resets the count.

How to Test

1. Focused approval, gateway, CLI, Telegram, relay and plugin-origin suites: 305 passed.
2. The new or changed tests fail on cb0ca31c.
3. ruff check passes on touched files.
4. Three gateway tests (test_automation_retry, test_hosted_mux_runtime, test_hosted_group_members) also fail on 99ffc621.

Checklist

Code

• My commit messages follow Conventional Commits
• My PR contains only changes related to this fix
• I've added tests for my changes

Documentation & Housekeeping

• I've updated relevant documentation — N/A
• I've updated cli-config.yaml.example if I added/changed config keys — N/A
• I've updated tool descriptions/schemas if I changed tool behavior — N/A

For New Skills

N/A

Screenshots / Logs

N/A: test runs above.

Refs RIG-8544

Co-authored-by: Matt Wilkinson matt@rigel.build

[linear-code]

RIG-8544

[rigel-mintaka]

Review round 1 (head cb0ca31c)

Full lens fan-out (correctness, security, design, style). The direct fixes hold: bulk approve skips DENY, session and always become once, DENY doesn't coalesce, and subagent auto-approve refuses DENY. The invariant still breaks in three ways, so this PR is not ready.

Severity	Finding	Disposition
High	Card taps and /approve resolve the oldest request in the session, not the one shown. A tap on an ordinary card can release a queued DENY.	Fixing: the resolver never releases a DENY unless its own request_id names it. The Telegram and relay cards will carry request_id.
High	Same root cause, through parallel tool calls and plain /approve	Fixing with the above
High	The interactive CLI hardcodes smart_denied=False, so the reviewer warning is never shown	Fixing: carry smart_denied through PendingControls and GatewayAgent
Medium	A cached session grant skips the reviewer	Fixing: skip the cache when review_first is set
Medium	The CLI panel can drop every command row, and it miscounts rows	Fixing: keep at least one command row; correct the count
Medium	Subagent auto-approve also matches display text; its docstring is wrong	Fixing: use the kwarg only; correct the docstring
Medium	Reviewer note string is duplicated; an over-length comment	Fixing
Medium	No end-to-end test for DENY queued ahead of an ordinary card	Adding it
Low	Assorted style and test nits	Cheap ones fixed; rest noted

Next: fixes with tests that fail first, then round 2.

[rigel-mintaka]

Round 1 fixes (head 6d66927e)

All round 1 high and medium findings are addressed in 6d66927e:

• Oldest-request resolution: a reply that doesn't name a DENY by its ID never releases it. Telegram and relay cards carry their request ID.
• CLI warning: smart_denied now reaches the interactive CLI.
• Cached grant: with review_first, the reviewer always runs.
• Panel: at least one command row stays visible, and the row count is fixed.
• Subagent check: it uses the kwarg only, and its docstring is corrected.
• Cleanup: one shared note constant; the long comment is shortened; the end-to-end mixed-queue case is added.

All 9 new or changed tests fail on cb0ca31c. Focused suites pass. Round 2, with every lens, is running on this head.