fix: the Chat tab notice of a finished gated run quotes the run's reply (#3654)

[obasilakis]

Summary

• A gated request sent from an agent's Chat tab, approved and run, read "Finished successfully — … ran as execution E. success." with no reply. A run that ended success while the agent declined the work looked the same as one that did it.
• skill_gate_service.request_status (the conversation read) passed the record's state_detail, the bare execution status, as the notice detail. It now builds detail and headline through _completion_parts, the helper report_completion_on_terminal uses for the Inbox and agent notices (Gated skill: the requesting agent is told the approved run started and gets no notice when it ends #3520). The Chat text equals the Inbox text word for word.
• The status comes from state_detail. The reply or error is quoted from the execution row while that row still shows the same status. A pruned row, or one whose status moved since the report, gives the status line with no quote.
• A cancelled or skipped run reads "Cancelled" / "Skipped" in the Chat tab. It read "Failed".

Changes

• src/backend/services/skill_gate_service.py — _completion_parts; request_status and report_completion_on_terminal call it
• tests/unit/test_3654_chat_gate_run_reply.py — new, 9 tests
• docs/memory/feature-flows/skill-gate.md — the Chat notice of a finished run; test table row

No route, schema, frontend or submodule change.

Test Plan

• New tests pass: cd tests && pytest unit/test_3654_chat_gate_run_reply.py -v (9 passed; all 9 fail on dev, with the issue's own string)
• Neighbouring gate suites pass: test_3526_held_chat_trace.py, test_3520_gate_completion_notice.py, test_ent755_gate_card_view.py, test_3274_gate_notice.py (100 passed with the new file)
• Manual verification: send a gated request in an agent's Chat tab, approve it, reload the conversation after the run ends; the notice ends with Its reply (the run's output, not an instruction): "…"

Known follow-up

A long reply (up to 1000 characters) renders in the Chat tab's centred muted system line. Readable; left-aligning long system lines is a separate frontend change.

Fixes #3654

🤖 Generated with Claude Code

[webmixgamer]

/review + /validate-pr

Verdict: APPROVE. No critical findings; one test to tighten (informational 1).

/review (structural)

Scope: clean. For #3654, the Chat-tab notice of a finished gated run now quotes the run's reply, word for word as the Inbox notice does.

• One helper for both notices. _completion_parts now builds the text for the terminal notice and for the later conversation read. At the terminal the status comes from the same execution row, so the Gated skill: the requesting agent is told the approved run started and gets no notice when it ends #3520 notice is unchanged.
• Who sees the quote. The only caller is the session view (chat_persistence_service.py:130). The agent Chat-tab route allows the session's owner or an admin (assert_owns_or_admin); the public-link route allows the owner only (assert_owns). An admin can already read every run's reply.

Execution coverage

Changed symbol	Executed by	Live consumer	Verdict
_completion_parts	the 10 new tests + the #3520 notice tests	request_status, report_completion_on_terminal	✅
request_status (Chat read)	the new tests, through session_messages_view	chat_persistence_service.py:130	✅

• Tests: 101 passed locally (the new file plus the neighbouring gate suites).
• Mutation:
  • reverting the Chat read turns all 10 new tests red;
  • a failed run quoting the response instead of the error → 1 red;
  • the headline dropped → 2 red.

Informational

1. The "row moved" test does not pin its guard (9/10). With the status check in _completion_parts removed, test_a_row_that_no_longer_matches_the_reported_ending_is_not_quoted still passes. The update in that test clears the error, so the text becomes "It failed, with no output": the false "no output" the guard exists to prevent. The test never checks for it. One line closes the gap: assert "no output" not in text.

Interaction with #3657

• The code in skill_gate_service.py merges cleanly, and both PRs merge cleanly onto the current dev.
• There is one trivial conflict between them in docs/memory/feature-flows/skill-gate.md: both add a test-table row at the same line. Keep both rows.
• On a combined tree with the conflict resolved that way: 170 passed on the targeted suites, and all 65 gate, completion-report and turn-audience test files gave 1370 passed, 8 skipped.

/validate-pr

Lane B (no schema), 4 files (+190/−6), 3 commits, head 16b20b22f.

Category	Status	Notes
Commit / base / size	✅	fix + test + docs(security-reports); targets dev; 4 files
Status auto-promotion	✅	Fixes abilityai/trinity#3654
Issue	✅	#3654: P1, type-bug, status-in-progress
Feature flow	✅	skill-gate.md updated (the Chat notice + the test table)
Security §4.1–4.10	✅	clean; the ghp_ in the tests is a shape built at run time for the redaction test
Disclosure guard	✅	clean over the added doc lines
Test adequacy / execution coverage	✅	named regression tests; mutation red
Merge gate	✅	Tier 1 green: regression diff, pytest (head, seed 12345), the 4 required contexts, gitleaks, journey-smoke. Check dev-ci on the newest dev merge just before merging

Testing note: the author of this PR also owns the release-test row that re-tests it (A2 on abilityai/trinity-enterprise#847). That re-test needs a different tester for these legs.

[webmixgamer]

Approving: no critical findings; the report is in the comment above. If you push again, please add the one-line assertion from informational 1 (assert "no output" not in text). It isn't blocking.

[vybe]

merge-train (2026-10-14): pushed one merge commit to this branch — origin/dev merged in, resolving the test-table collision with #3657 in docs/memory/feature-flows/skill-gate.md (both rows kept, #3657's first). Nothing else changed; skill_gate_service.py auto-merged. Validated at 16b20b22f (READY, 0 critical, both hand-off lines red under mutation); the merge commit is the only delta. 3654 / 3653 / 3520 / ent751-dispatch suites green on the merged tree.