fix(guards): C4 backfill slice -- 6 confirmed guard/ownership findings (t_9c9e757c)

[ang-fleet-workers]

C4 backfill slice (FleetReview retro-backfill 2026-09-27, card t_9c9e757c, parent t_a2bf7813). All 14 instances re-checked at 5a9d284.

Fixed (6). Each has a test that fails on base and passes with the fix

Finding	File	Fix
#813	agent/quota_registry_gate.py	A missing or unparseable observed_at now fails open, as the docstring promises
#1019	agent/lsp/manager.py	An OSError on the slot dir means LSP is skipped. It no longer raises out of enabled_for()
#1026	cron/scheduler.py	The deadman exemption now covers only the host that deadman owns
#1229	scripts/ci/live_comment.py	A lookup or update outage no longer kills the poller. The post is retried on the next poll
#1230	plugins_cmd.py	The pinned-plugin remedy shell-quotes the source, in both the CLI and the dashboard
#1238	tools/mixture_of_agents_tool.py	Reference responses moved out of the system message into a tagged data block in the user message

Not fixed here (8)

• fix(lifecycle-guard): gate the read-only Python path mask on a whole-body allowlist #1017 cron/lifecycle_guard.py:2051, superseded. fix(lifecycle-guard): mask allowlist refuses module callables as values + callback keywords (backfill C1) #1348 (6662bbd) landed a broader fix on main: an imported name may only be loaded as the callee of an allowlisted call. This slice's 5 callback cases pass against it, so the edit and its test were dropped when rebasing onto 086b4a0.
• revert(budget): drop grace-turn side-effect lockout (#37) #1145 tool_executor.py:783, by-design. revert(budget): drop grace-turn side-effect lockout (#37) #1145 is itself the deliberate revert of the grace-turn lockout (fork-PR audit DROP: 0 denials in 555,617 tool calls).
• revert(mem0): drop #251 correction-marker salience-gate bypass (fork-PR audit DROP) #1213 capture_drain.py:346 and capture_router.py:9, by-design. revert(mem0): drop #251 correction-marker salience-gate bypass (fork-PR audit DROP) #1213 is the deliberate revert of the mem0: bypass salience gate for corrections #251 correction bypass (audit FINAL DROP, card t_ff91a776).
• audit(cron_tools): batched fork-PR audit reverts — 2 PRs (#1172, #1173) #1228 vision_tools.py:2199, by-design. This is the audited revert of the feat(video_analyze): SSRF-filtering egress proxy for yt-dlp (default OFF) #211 SSRF proxy. The proxy was never enabled in any config, so exposure does not change.
• feat(fallback): fallback_events ledger in blackbox (fallback-cache spec Phase 1) #1211 agent_runtime_helpers.py:1891, false positive. record_restore_refused already swallows every error: try: ... except Exception: logger.debug("restore_refused ledger write failed", ...). It cannot raise past the return False.
• ci: Blacksmith as the paid third venue rung in the slice generator (t_6804be8c) #1147 run_tests_parallel.py:605, false positive / by-design. merge_group code is already approved for merge and runs at push-to-main trust. The pull_request fork case is gated (same_repo == "true").
• fix(ci-overflow): admit hosted jobs at measured p90 estimate + headroom M; release runner-less cancels (t_f459aa52) #1216 ci_overflow_ledger.py:286, dropped, wrong class. This is a budget-allocation limitation of the admission estimator, not an authz or guard bypass.

Verification

• Local runs were narrow, via test-gate. The new tests failed on base (13 failing test cases) and all pass with the fix.
• Neighbouring suites also pass: lifecycle_guard x4, host_down_gate, quota wiring, plugins_cmd, fallback_policy, fork toolsets. 44+22+24 and 95+54+167 passed.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[ang-fleet-lander]

🤖 merged-by: apollo · lane: kanban-merge-pass · gate: ADVISORY (FleetReview not green for 1af0a00): fleetreview-advisory-20260927-standing.md · why: t_9c9e757c: C4 backfill slice: hermes-agent agent/tools/cron/scripts guards — 14 instances; Argus off card review (Ace 13:08), CI green

[Kyzcreig]

🤖 merged-by: apollo · lane: review · gate: ADVISORY (FleetReview not green for 1af0a00): fleetreview-advisory-20260927-standing.md · why: t_9c9e757c: CI green; Apollo review pass 16:40

[ang-prism]

FleetReview

Review: post-merge · head ab6b6d46747e · duration 8m 03s
Profile: light (merit: default light: lines 264<800, files 12<1000000, hunks 18<1000000, no hot path) · policy: below-size-and-path-gates
Roster: B-state → gpt-6-sol (openai), C-assert-xhigh → claude-code-opus-5-5 (anthropic), G → grok-4.6 (xai), L6 → gpt-6-sol (openai)

Post-merge review (fleetreview:post-merge override): this reviewed the merge commit against its first parent — the bytes that already shipped. It is not a pre-merge gate pass.

profile: light (rule: default light: lines 264<800, files 12<1000000, hunks 18<1000000, no hot path) · round 0 · members: B-state, L6, C-assert-xhigh, G · families: anthropic,openai,xai

Confidence: 3/5

Findings

• P1 cron/scheduler.py:610 — Gate bypass · agreed: B-state (openai)
• P1 cron/scheduler.py:3609 — Gate Bypass · agreed: L6 (openai)
• P1 tools/mixture_of_agents_tool.py:93 — Reference responses are not escaped, so a response can close the <reference_responses> tag and forge a 'User query:' section · agreed: C-assert-xhigh,G (anthropic, xai)

---

FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, D=grok-4.6 · cost: $1.57 · duration: 7m 59s · rounds: 1 · files examined: 12